- Email: [email protected]
DeltaT50—a new method to assess temporal ventricular repolarization variability
Available online at www.sciencedirect.com
Journal of Electrocardiology 44 (2011) 477.e1 – 477.e9 www.jecgonline.com
DeltaT50—a new method to assess temporal ventricular repolarization variability Christina Abrahamsson, PhD,⁎ Corina Dota, MD, Bo Skallefell, BSc, Leif Carlsson, PhD, Dunia Halawani, MSc, Lars Frison, PhD, Anders Berggren, MD, PhD, Nils Edvardsson, MD, PhD, Göran Duker, PhD AstraZeneca R&D, Mölndal and Sahlgrenska Academy at Sahlgrenska University Hospital, Göteborg, Sweden Received 16 December 2010
Abstract
Background: Increased beat-to-beat variability in cardiac repolarization time is a tentative risk marker of drug-induced torsades de pointes. We developed a new, automatic method based on the temporal variability of the T-wave down slope to assess this variability. Method and Results: Leads V1 to V6 of resting electrocardiograms were recorded in 42 healthy subjects (18-68 years, 22 men). The temporal variability at 50% of the T-wave down slope, deltaT50 (1.5 ± 0.41 milliseconds; range, 0.86-2.66 milliseconds), was measured with an accuracy of 1 millisecond on at least 9 pairs of electrocardiogram complexes with a signal-to-noise ratio more than 10 and changes in the R-R interval less than 150 milliseconds. The correlation between repeated measurements of deltaT50 was high. DeltaT50 was measured without corrections for age, sex, heart rate, T-wave amplitude, signal-to-noise ratio, R-R variability, and QTcF because none of these factors explained more than 4% of the within-subject deltaT50 variability. Conclusion: The beat-to-beat repolarization variability was measured with high fidelity with the deltaT50 method and was a robust measure in healthy volunteers. © 2011 Elsevier Inc. All rights reserved.
Keywords:
Beat-to-beat; QT variability; Repolarization instability; Automatic analysis
Introduction Variability in the cardiac repolarization time may indicate an electrical instability that constitutes a potential risk of lifethreatening arrhythmias.1 Both increased variability and prolongation of the cardiac repolarization time are considered to be indicators of a reduced repolarization reserve, which may imply an increased risk of torsades de pointes tachycardia (TdP).2,3 However, the 2 variables reflect different aspects of a reduced repolarization reserve because increased variability may occur at any QT interval duration,4 whereas drug-induced QT prolongation does not necessarily lead to an increased QT variability and TdP.5 In experimental studies, an increased beat-to-beat variability of the QT interval or of the monophasic action potential duration was superior to QT prolongation in predicting the risk of imminent TdP or sudden cardiac death.5-8 The potential value of QT variability as a diagnostic marker for TdP was ⁎ Corresponding author. AstraZeneca R&D, Mölndal and Sahlgrenska Academy at Sahlgrenska University Hospital, 431 83 Mölndal, Sweden. E-mail address: [email protected] 0022-0736/$ – see front matter © 2011 Elsevier Inc. All rights reserved. doi:10.1016/j.jelectrocard.2011.04.010
demonstrated in patients with congenital long-QT syndrome (LQTS), where combined measurements of the QT variability and the QT interval improved the identification of mutation carriers as compared with QT interval measurements alone.9 Hence, increased beat-to-beat repolarization variability may be a valuable risk marker for drug-induced proarrhythmia and a potential tool for identifying patients at increased risk of such proarrhythmia, but would require a precise and accurate method that automatically determines the cardiac repolarization time on a beat-to-beat basis. Methods that are available for assessing beat-to-beat repolarization variability include algorithms such as the tangent method,10 template matching,11 or measurement to the maximum derivative of the T-wave down slope.12 Initial work to measure beat-to-beat QT variability using the tangent method in EClysis, a computerized system for electrocardiogram (ECG) analysis, 10 showed that this algorithm worked appropriately in ECG recordings that were of good quality and had a normal QRST morphology, whereas a more robust method seemed to be needed to enable analysis of less perfect ECG signals on a single-beat basis. The idea of measuring the repolarization variability along the T-wave down slope
477.e2
C. Abrahamsson et al. / Journal of Electrocardiology 44 (2011) 477.e1–477.e9
emerged after superimposing a large number of consecutive ECG complexes from one ECG lead, as the T-wave down slope appeared to shift in parallel when the QT interval varied. Thus, to enable automatic measurement of the QT interval variability, a novel algorithm, deltaT50, which measures the temporal variability of the steep part of the T-wave down slope instead of the end of the T wave, was developed. Furthermore, to allow for automatic analysis of a large number of ECG complexes, the new algorithm was incorporated into a special version of EClysis used for research purposes. This study describes the rationale and development of the deltaT50 method in young, healthy subjects and presents initial results of application of the method in studies in healthy men and women of different ages.
Methods All subjects gave their written informed consent to participate in the study. The study protocols were approved by the ethics committee of Gothenburg University. ECG recording and definitions of measured variables Precordial ECG leads V1 to V6 were recorded during 10 minutes of supine rest in the morning after a 1 night fast. The ECG signals were sampled at 500 Hz and imported to EClysis.10 The EClysis system includes software that processes digitally recorded ECG signals from up to 12 leads at 500 Hz, using strictly defined algorithms to detect the PQRSTU points and to measure ECG intervals and amplitudes. The interday and intraday variability in major ECG intervals and amplitudes in healthy men and women was determined.13 The high-precision ECG measurements were validated by showing a high level of agreement when tested against reference values from a variety of ECGs from patients in the Common Standards for Quantitative Electrocardiography database.14 Hence, the deltaT50 algorithm was added to this validated version of EClysis, and the following calculations were done: on each single ECG complex, the intervals from the intersection between the RS line and the isoelectric line to 20%, 50%, and 80% of the T-wave down slope were defined as the T20, T50, and T80 intervals, respectively (Fig. 1A, C). If the RS line did not cross the isoelectric line, as in Fig. 1A, the interval started from the intersection between the extrapolated RS line (ie, the tangent of the RS line) and the isoelectric line. The averages of the absolute values of the beat-to-beat changes in the T20, T50, and T80 intervals were denoted deltaT20, deltaT50, and deltaT80, respectively (Fig. 1D). Accordingly, the average beat-to-beat change in the R-R intervals was denoted deltaRR (Fig. 1D). To reduce noise artifacts, the amplitude of each sample of the T-wave down slope was recalculated from the moving average of the amplitude of 21 consecutive samples, 10 before and 10 after the current sample (Fig. 1B).15 The signal-to-noise ratio (SNR) was calculated for each beat as described in the following. The signal amplitude was derived from the average T-wave amplitude during the interval between 50 milliseconds after the J point and the end of the
T wave. The noise amplitude was calculated as the average absolute difference in amplitude between the smoothed and the original curve in the diastolic interval from the end of the T wave to the start of the P wave. The SNR was subsequently calculated by dividing the signal amplitude by the noise amplitude. Following this procedure, the deltaT50 measurements were performed on a strict beat-to-beat basis. The deltaT50 method Using ECG recordings from 10 healthy subjects (6 men) aged 21 to 29 years, the deltaT50 method was developed and validated on the basis of the following 4 steps: (1) Measurements of deltaT20, deltaT50, and deltaT80 before and after smoothing. DeltaT20, deltaT50, and deltaT80 were calculated before and after smoothing on the first 30 beats each minute of each 10-minute ECG recording, that is, on 290 beat pairs/subject. (2) Investigation of the relationship observed between beat-to-beat changes in the T50 and R-R intervals. The correlation between the beat-to-beat changes in the T50 and the preceding R-R intervals was investigated on the first 30 beats of each minute of the 10-minute ECG recording in each subject, that is, on 290 beat pairs/subject, by applying linear regression to the data. (3) Effects of added noise to the ECG signals on deltaT20, deltaT50, and deltaT80. The first 30 beats of each ECG recording were analyzed to examine the influence of added noise and set criteria for acceptable signals. Beat pairs where the preceding R-R interval changed more than 150 milliseconds were excluded, which resulted in 27 to 29 beat pairs/ subject. Random white or 50-Hz sinusoidal noise of increasing amplitude (10, 50, 100, and 200 μV) was added to the ECG signals by means of a function in EClysis. The average deltaT20, deltaT50, and deltaT80 were calculated at each noise level before and after 21 p smoothing. The SNR was calculated before the smoothing. (4) Investigation of the least number of beats and the acceptable SNR for an estimation of deltaT50 with a precision of ±1 millisecond. To define the criteria for acceptable signals, that is, the least number of beat pairs and the acceptable noise level for an estimation of deltaT50 with a precision of ± 1 millisecond, a systematic investigation was undertaken in each subject. DeltaT50 was calculated at each noise level on a reduced number of beats (beats 1-5, 1-10, and 1-20) and compared with the deltaT50, resulting from the analysis of all 30 beats in the absence of noise.
DeltaT50 in healthy subjects at rest In addition to the 10 subjects who participated in the first part of the study described above, another 32 healthy subjects (16 men) aged 18 to 68 years were included to
C. Abrahamsson et al. / Journal of Electrocardiology 44 (2011) 477.e1–477.e9
477.e3
Fig. 1. A, An example of an ECG recording from one of the subjects in the study, illustrating the preceding R-R intervals and T50 intervals for a beat pair, beats 1 and 2. The R-R interval was the time between the R peaks, and the T50 interval was the time from the intersection between the RS line and the isoelectric line to 50% of the T-wave down slope (C). B, The original curve (black) was smoothed with 21 points (smoothed curve in gray) to reduce noise artifacts; that is, the amplitude of each sample was recalculated from the average amplitude of the preceding and following 10 samples together with the current sample. As an example, sample number 50 of the smoothed curve, marked with the curved arrow, is the average of samples 40 to 60 of the original curve, marked with the bracket. C, The time from the intersection of the RS line and the isoelectric line to 20%, 50%, and 80% of the T-wave down slope, the T20, T50, and T80 intervals, was measured on the ECG signals before and after smoothing. Interpolation between the discrete samples was used to determine the respective interval. C, The different levels marked on the smoothed curve, which is also shown in panel B. D, The definition of deltaT50 and deltaRR; that is, deltaT50 was the average change in the T50 intervals of consecutive beat pairs, and deltaRR was the average change in the preceding R-R intervals of consecutive beat pairs.
investigate the role of age, sex, SNR, T-wave amplitude, QTcF, heart rate (HR), and R-R variability (measured as deltaRR) for the short- and long-term variability of deltaT50. DeltaT50 was measured once every minute on 30 beats per 10 minutes on 2 occasions in each subject. In the first 10 subjects, the ECGs were recorded on 2 different days at an interval of 1 to 3 weeks; in the subsequent 32 subjects, ECGs were recorded on the same day but separated by 1 hour. The first 30 beats every minute were used for the calculation of deltaT50, and beat pairs where the preceding R-R interval changed more than 150 milliseconds or the SNR was less than 10 were excluded. This resulted in 9 to 29 beat pairs for the calculation of deltaT50. DeltaT50 was measured on both leads V2 and V5 in the 10 subjects and on either lead V5 or lead V2 in the 32 subjects. The HR and deltaRR were measured on all 30 beats. QTcF was automatically measured
on the average ECG complex of all 30 beats with the tangent method in EClysis.10
Statistical methods Statistical analyses were primarily performed using repeated-measurements analysis of variance, accounting for correlated observations for the subjects within and between the 2 study sessions. Analyses addressing subject factors, being based on the mean levels for each subject, for example, evaluating the potential relationship between deltaT50 and other variables of interest, were based on analysis of variance or linear regression, depending on whether only continuous or also categorical variables were included. P values less than .05 were regarded as statistically significant.
477.e4
C. Abrahamsson et al. / Journal of Electrocardiology 44 (2011) 477.e1–477.e9
Results The deltaT50 method (1) Measurements of deltaT20, deltaT50, and deltaT80 before and after smoothing. The effect of 21 p smoothing on deltaT20, deltaT50, and deltaT80 calculated on the first 30 beats each minute of the 10minute ECG recordings in the 10 subjects is depicted in Fig. 2. Smoothing decreased deltaT20, deltaT50, and deltaT80 significantly (P b .0001), but as seen in Fig. 2, the decrease was very small. The average deltaT20, deltaT50, and deltaT80 in the 10 subjects were 1.63 ± 0.69, 1.63 ± 0.71, and 1.84 ± 0.71 milliseconds before smoothing and 1.56 ± 0.69, 1.57 ± 0.70, and 1.62 ± 0.69 milliseconds after 21 p smoothing (Fig. 3A). (2) Investigation of the observed relationship between beat-to-beat changes in the T50 and R-R intervals. There was a positive and linear correlation between the beat-to-beat changes in the T50 and the R-R intervals in all subjects, and the average slope of the regression lines was 0.0178 (range, 0.0112-0.0247). However, the range of the beat-to-beat changes in the R-R interval varied substantially between subjects, between ±100 and ±500 milliseconds, as illustrated in Fig. 3B. Hence, to compare deltaT50 at a similar range of R-R interval changes in all subjects, we calculated deltaT20, deltaT50, and deltaT80 only on beat pairs with a change in the preceding R-R interval of less than 150 milliseconds (marked with b in Fig. 3B). Subsequently, when the relationship between deltaT50 and deltaRR was investigated, only 19% of
the variability in deltaT50 could be explained by variations in deltaRR (Fig. 6D; Table 1), and we decided not to correct deltaT50 for differences in deltaRR. With this criterion, deltaT20, deltaT50, and deltaT80 after smoothing were 1.30 ± 0.49, 1.28 ± 0.42, and 1.34 ± 0.36 milliseconds, respectively (Fig. 3A), and the between- and within-subject variability in deltaT50 decreased (Fig. 3C). (3) Effects of added noise to the ECG signals on deltaT20, deltaT50, and deltaT80. The original ECG signals were generally of good quality, and the average SNR was 87 (range, 40-169). After the addition of 10, 50, 100, and 200 μV of random white or 50-Hz sinusoidal noise, the average SNR decreased to 64, 21, 11, and 8 at the respective noise level. Because of the noise artifacts, deltaT20, deltaT50, and deltaT80 increased with decreasing SNR. Before smoothing, the increase became statistically significant when the average SNR was 21 or less (Fig. 4). DeltaT50 increased less than deltaT20 and deltaT80, implying that deltaT50 was the variable least sensitive to noise. 21 p smoothing reduced the noise artifacts, restored the morphology of the ECG signals (Fig. 4), and abolished the increase in deltaT20, deltaT50, and deltaT80 at an average SNR of 21 (Fig. 4). However, smoothing of the ECG signals was not sufficient to abolish the influence of noise on deltaT80 when the average SNR before smoothing was 11 or less, and on deltaT20 and deltaT50 when the average SNR was 6. Hence, deltaT80 was more sensitive to noise than deltaT20 and deltaT50. (4) Investigation of the least number of beats and the acceptable SNR for an estimation of deltaT50 with a precision of ±1 millisecond. The deltaT50 calculated on a reduced number of beats in the individual subjects showed that deltaT50 calculated on beats 1 to 10 (8 beat pairs in 1 subject and 9 beat pairs in the other 9 subjects) and 1 to 20 (17-19 beat pairs) never differed more than 1 millisecond in any subject from deltaT50 calculated on beats 1 to 30 (27-29 beat pairs) when the SNR was 10 or higher. Hence, the least number of beat pairs required was 9, and the acceptable SNR was 10 for an estimation of deltaT50 with a precision of ±1 millisecond. DeltaT50 in healthy subjects at rest
Fig. 2. DeltaT20, deltaT50, and deltaT80 were calculated on the first 30 beats each minute of the 10-minute ECG recording in the 10 subjects on the original curves (no smoothing) and after 21 points smoothing (for an explanation of the variables and the smoothing procedure, see Fig. 1). The correlation coefficients (r) for the relationship between deltaT20, deltaT50, and deltaT80 measured after smoothing and plotted as a function of the corresponding measures before smoothing are shown in the legend in the figure. The gray line represents the line of identity.
The average deltaT50 was 1.5 ± 0.41 milliseconds in the 42 subjects, ranging from 0.86 to 2.84 milliseconds (Fig. 5A), and deltaT50 measured on the 2 occasions in each subject correlated well (Fig. 5A). Furthermore, the variability of deltaT50 over the 10-minute period in each subject was small, and deltaT50 measured during the first minute of the ECG recording correlated well with the average deltaT50 measured over the entire 10-minute period (Fig. 5B). The results of the linear regression analysis of the correlation between deltaT50 and HR, T-wave amplitude, SNR, deltaRR, and QTcF, respectively, are summarized in
C. Abrahamsson et al. / Journal of Electrocardiology 44 (2011) 477.e1–477.e9
477.e5
Fig. 3. A, The average deltaT20, deltaT50, and deltaT80 in 10 young, healthy subjects calculated on the first 29 beat pairs each minute of a 10-minute ECG recording, before smoothing (bars to the left), after 21 points smoothing (bars in the middle, marked with a), and only on the beat pairs where the change in the preceding R-R interval was less than 150 milliseconds (bars to the right, marked with b). Mean ± SEM. B. The change in the T50 interval of the 290 beat pairs (T50n − T50(n − 1)) described in panel A vs the change in the preceding R-R interval (R-Rn − R-R(n − 1)) for the same beat pairs (1 symbol per subject). All beat pairs were used to calculate the average deltaT20, deltaT50, and deltaT80 shown in the bars in the middle, marked with a, in panel A. Only the beat pairs where the absolute change in the preceding R-R interval was less than 150 milliseconds were used to calculate the average deltaT20, deltaT50, and deltaT80 shown in the bars to the right, marked with b, in panel A. C, Individual values of deltaT50 in the 10 young subjects each minute of a 10-minute ECG recording (time in minutes on the x-axis). The beats marked with a in panel C were used to calculate deltaT50 in the left graph, and the beats marked with b in panel C were used to calculate deltaT50 in the right graph.
Table 1. Within subjects, deltaT50 correlated significantly with all variables except QTcF, but the R2 values were low (range, 0.5-3.5%), and the influence of each variable on the variability of deltaT50 was negligible (Table 1). Between
Table 1 Linear regression was used to investigate the correlation between deltaT50 and HR, T-wave amplitude, SNR, deltaRR, and QTcF within and between the 42 subjects during 10 minutes of rest DeltaT50 vs:
HR T-wave amplitude SNR DeltaRR QTcF
Within subjects
subjects, deltaT50 correlated significantly only with HR and deltaRR (Fig. 6A-E), but again, the R2 values were low, and variations in HR and deltaRR explained only 10% and 19%, respectively, of the total deltaT50 variability (Table 1). As a comparison, 47% and 52% of the within- and betweensubject variability in the QT interval, respectively, was explained by variations in the HR. Finally, there was no correlation between deltaT50 and age or sex, and deltaT50 measured from lead V2 differed less than 1 millisecond from deltaT50 measured from lead V5 (Fig. 7).
Between subjects
P
R2 (%)
P
R2 (%)
Discussion
b.0001 .018 .041 b.0001 NS
2.6 0.7 0.5 3.5
.044 NS NS .004 NS
10
The parallel nature of the shift of the T-wave down slope between consecutive beats, which is visible when consecutive ECG complexes are superimposed, was demonstrated by almost identical deltaT20, deltaT50, and deltaT80 values. This fact, together with the notion that both deltaT20 and deltaT80 were marginally more sensitive to the effects of
19
The P and the R2 values for the respective correlations are shown. NS indicate nonsignificant.
477.e6
C. Abrahamsson et al. / Journal of Electrocardiology 44 (2011) 477.e1–477.e9
Fig. 4. The average deltaT20, deltaT50, and deltaT80 calculated on 27 to 29 beat pairs in the 10 subjects before and after smoothing and before and after added noise (random or sinusoidal noise). Only beat pairs with a change in the preceding R-R interval of less than 150 milliseconds were included. Values are expressed as mean ± SEM. Inserted are also representative recordings of the T wave from 30 beats from 1 subject, showing the effects of smoothing on the signal quality.
Fig. 5. A, The average deltaT50 for 10 minutes was measured on 2 occasions separated by 1 to 3 weeks in 10 subjects and by 1 hour in 32 subjects (R2 = 52%). B, DeltaT50 was stable for 10 minutes, and deltaT50 measured the first of the 10 minutes correlated well with the average deltaT50 of the 10 minutes (R2 = 75%).
added noise to the ECG signals, made us choose deltaT50 as the primary measure of the beat-to-beat variability in cardiac repolarization time. Compared with the presently used measures, which define the QT interval as the interval between the start of the QRS complex to the end of the T wave, the fiducial points of the T50 interval are more easily defined. The start of the T50 interval is set to the crossing point between the RS line and the isoelectric line and the end to 50% of the T-wave down slope. Thereby, variations in the width of the QRS complex are largely avoided as well as problems with determination of the end of the T wave. Initially, all beats were included in the calculation of the deltaT50 for a series of 30 beats. Not surprisingly, we found that a large alteration in the R-R interval was associated with a major change in the QT interval of the following beat. Pronounced R-R and QT variability was found in some, but not all, subjects, and the analysis was refined to exclude beats after a change in the preceding R-R interval of more than 150 milliseconds to calculate delta T50 at a similar range of R-R interval changes in all subjects. Noise artifacts usually disturb the analysis of single beats, but the 21 p smoothing, which had a negligible effect on deltaT50 measured on “clean” signals, efficiently removed the noise artifacts on ECG signals with an SNR above 10 and the error of the estimate of deltaT50 caused by these introduced artifacts. Beats with an SNR of less than 10 were therefore excluded from the calculation of deltaT50. With these exclusion criteria, 9 beat pairs were found to be sufficient to estimate deltaT50 with an accuracy of 1 millisecond; that is, values obtained from 9 pairs never differed by more than 1 millisecond as compared with values derived from 29 pairs in any individual subject. DeltaT50 was generally very low, on average, approximately 1.5 milliseconds. Studies of beat-to-beat QT variability using comparable or similar methods are scarce in the
C. Abrahamsson et al. / Journal of Electrocardiology 44 (2011) 477.e1–477.e9
477.e7
Fig. 6. The correlations between deltaT50 and HR (A), T-wave amplitude (B), SNR (C), deltaRR (D), and QTcF (E). The P and R2 values for the respective correlation are summarized in Table 1.
literature (summarized in Pueyo et al16). The root mean square of differences between consecutive QT intervals (measured from Q start to T end), that is, similar to the deltaT50 in the present study, was automatically analyzed from several hours of ambulatory ECGs in healthy subjects and patients with LQTS.17 In that study, the average root mean square of differences between consecutive QT intervals was approximately 7 milliseconds in the healthy subjects and
Fig. 7. DeltaT50 was measured on leads V2 and V5 in 10 subjects once every minute during 10 minutes of rest (10 values per subject, 1 symbol per subject). The dotted line represents the line of identity. The difference in deltaT50 measured on the 2 leads was generally very small (b1 millisecond) but consistently higher or lower at all time points in the same subject.
15 milliseconds in the LQTS patients.17 Hinterseer et al4,9,18 manually measured 30 consecutive QT intervals from ECG printouts, and the short time variability, defined as the average beat-to-beat change in the QT interval divided by the square root of 2, was approximately 4 milliseconds in their control groups. Couderc et al19 used a wavelet transformation methodology and calculated the beat-to-beat shift of the resulting waveforms constituting the T wave and found a repolarization variability of 2 ± 4 milliseconds in healthy subjects. The deltaT50 variability over time was low, and the reproducibility of the measurement when carried out 1 hour or 1 to 3 weeks apart in the same individual was good. Moreover, the deltaT50 calculated on the first minute of the 10-minute continuous ECG recording correlated well with the average deltaT50 for the entire recording period in all subjects. Consequently, the individual deltaT50 may be determined with appropriate accuracy at a single time point and may be monitored minute by minute. By excluding beats after pronounced changes in the R-R interval, only 19% of the total variability in deltaT50 between the subjects and only 3.5% within the subjects could be explained by the R-R variability. There was thus no cause for correcting deltaT50 for deltaRR. The lack of correlation between deltaT50 and QTcF emphasizes that the 2 measures of cardiac repolarization ability are unconnected. In fact, deltaT50 was measured without correction for any of the other tentatively confounding factors,20 HR, T-wave amplitude, and SNR, that varied between the subjects or within the same subject over time. Considering that measurements of the QT interval involve rate correction formulae that overcorrect and undercorrect values deviating
477.e8
C. Abrahamsson et al. / Journal of Electrocardiology 44 (2011) 477.e1–477.e9
from the HR of 60 beats/min, this is a great advantage. The choice of ECG lead (V2 or V5) was also found to be of minor importance as long as the measurements over time were based on recordings from the same lead. A risk marker for proarrhythmias should ideally have a very high specificity and be measured with high precision in an automatic and operator-independent manner. With the deltaT50 method, it is possible to determine the beat-to-beat repolarization variability automatically with high precision on only 9 beat pairs, and the method was found to be robust and required no correction for variations in factors such as age, sex, QTcF, T-wave amplitude, SNR, or time dependence. The stability of deltaT50 when followed over time enables detection of small deviations from the baseline value. The significance of beat-to-beat repolarization variability as a risk marker is under evaluation, and ECG databases, such as those in the Telemetric and Holter ECG Warehouse, 21 may provide opportunities to test new methods in ECGs of subjects, patients on various pharmacologic agents, and patients who have experienced spontaneous or drug-induced ventricular arrhythmias.
Limitations Excluding beats after large changes in the R-R interval was a way to standardize the measurements, and we cannot rule out that the variability in repolarization after the largest changes in the R-R interval may contain important information to be considered in future studies with deltaT50. For example, after treatment with ibutilide, the QT variability index increased only during random interval right atrial pacing and not during sinus rhythm, possibly implying that measurements at baseline alone may not always disclose prevalent abnormal phenomena.22 The relationship between beat-to-beat changes in the T50 and R-R intervals should therefore be characterized for any study population other than healthy volunteers. The primary objective of the present study was to describe the development and the validation of a novel methodology to assess temporal variability of ventricular repolarization. To this end, we applied the technique on digital ECGs recorded from healthy subjects studied under standardized conditions. However, for deltaT50 to become a clinically useful tool, it must also work reliably in patients with apparently normal or abnormal ECGs as a consequence of, for example, disease or medication. Evaluation of the methodology in such patient populations is currently ongoing.
Conclusion The beat-to-beat repolarization variability in healthy subjects was automatically measured with the deltaT50 method with an accuracy of 1 millisecond on at least 9 beat pairs with an SNR more than 10 and changes in the preceding R-R interval less than150 milliseconds. With these criteria, the variability of deltaT50 over time was low, and
correction for differences in age, sex, HR, T-wave amplitude, SNR, deltaRR, and QTcF was not needed. Acknowledgments We are indebted to Lars Lind, MD, PhD, and Catherine Åhlund, RN, for their contribution to the studies from which data were collected. We are also grateful to John Gustafsson, PhD, for fruitful input and discussions during the initial development of the method. References 1. Berger RD. QT variability. J Electrocardiol 2003;36(Suppl):83. 2. Thomsen MB, Matz J, Volders PGA, Vos MA. Assessing the proarrhythmic potential of drugs: current status of models and surrogate parameters of torsades de pointes arrhythmias. Pharmacol Therapeut 2006;112:150. 3. Roden DM. Long QT syndrome: reduced repolarization reserve and the genetic link. J Int Medicine 2006;259:59. 4. Hinterseer M, Thomsen MB, Beckman B-M, et al. Beat-to-beat variability of QT intervals is increased in patients with drug-induced long-QT syndrome: a case control pilot study. Eur Heart J 2008;29:185. 5. Carlsson L, Andersson B, Linhardt G, Löfberg L. Assessment of the ion channel-blocking profile of the novel combined ion channel blocker AZD1305 and its proarrhythmic potential versus dofetilide in the methoxamine-sensitized rabbit in vivo. J Cardiovasc Pharmacol 2009;54:82. 6. Thomsen MB, Verduyn SC, Stengl M, et al. Increased short-term variability of repolarization predicts D-sotalol–induced torsades de pointes in dogs. Circulation 2004;110:2453. 7. Thomsen MB, Truin M, van Opstal JM, et al. Sudden cardiac death in dogs with remodelled hearts is associated with larger beat-to-beat variability of repolarization. Basic Res Cardiol 2005;100:279. 8. Thomsen MB, Volders PGA, Beekman JDM, Matz J, Vos MA. Beat-tobeat variability of repolarization determines proarrhythmic outcome in dogs susceptible to drug-induced torsades de pointes. J Am Coll Cardiol 2006;48:1268. 9. Hinterseer M, Beckman B-M, Thomsen MB, et al. Relation of increased short-term variability of QT interval to congenital long-QT syndrome. Am J Cardiol 2009;103:1244. 10. Dota C, Skallefell B, Edvardsson N, Fager G. Computer-based analysis of dynamic QT changes: toward high precision and individual rate correction. Ann Noninvasive Electrocardiol 2002;7:289. 11. Berger RD, Kasper EK, Baughman KL, Marban E, Calkins H, Tomaselli GF. Beat-to-beat QT interval variability. Novel evidence for repolarization lability in ischemic and nonischemic dilated cardiomyopathy. Circulation 1997;96:1557. 12. Mortara DW. Automated QT measurement and application to detection of moxifloxacin-induced changes. Ann Noninvasive Electrocardiol 2008;14(Suppl 1):S30. 13. Dota CD, Edvardsson N, Skallefell B, Fager G. PC-based ECG waveform recognition-validation of novel software against a reference ECG database. Ann Noninvasive Electrocardiol 2009;14(Suppl 1):S42. 14. Dota CD, Edvardsson N, Schützer KM, et al. Inter- and intraday variability in major electrocardiogram intervals and amplitudes in healthy men and women. PACE 2003;26:361. 15. Bracewell R. Noise reduction by smoothing. In: Bracewell R, editor. Fourier analysis and imaging. New York: Kluwer Academic/Plenum Publishers; 2003. p. 269. 16. Pueyo E, Martinez JP, Laguna P. Cardiac repolarization analysis using the surface electrocardiogram. Phil Trans R Soc A 2009;367:213. 17. Němec J, Buncová M, Shusterman V, Winter B, Shen W-K, Ackerman MJ. QT interval variability and adaption to heart rate changes in patients with long QT syndrome. PACE 2009;32:72. 18. Hinterseer M, Beckmann B-M, Thomsen MB, et al. Usefulness of shortterm variability of QT intervals as a predictor for electrical remodelling and proarrhythmia in patients with nonischemic heart failure. Am J Cardiol 2010;106:216.
C. Abrahamsson et al. / Journal of Electrocardiology 44 (2011) 477.e1–477.e9 19. Couderc J-P, Zareba W, Burattini L, Moss AJ. Beat-to-beat repolarization variability in LQTS patients with the SCN5A sodium channel gene mutation. PACE 1999;22:1581. 20. Couderc J-P. Cardiac regulation and electrocardiographic factors contributing to the measurement of repolarization variability. J Electrocardiol 2009;42:494.
477.e9
21. Couderc J-P, Zareba W. An update on QT measurements and interpretation methodologies. Ann Noninvasive Electrocardiol 2008;14 (Suppl 1):S1. 22. Cheng A, Dalal D, Fetics BJ, et al. Ibutilide-induced changes in temporal lability of ventricular repolarization in patients with and without structural heart disease. J Cardiovascular Electrophysiol 2009;8:873.
Journal of Electrocardiology 44 (2011) 477.e1 – 477.e9 www.jecgonline.com
DeltaT50—a new method to assess temporal ventricular repolarization variability Christina Abrahamsson, PhD,⁎ Corina Dota, MD, Bo Skallefell, BSc, Leif Carlsson, PhD, Dunia Halawani, MSc, Lars Frison, PhD, Anders Berggren, MD, PhD, Nils Edvardsson, MD, PhD, Göran Duker, PhD AstraZeneca R&D, Mölndal and Sahlgrenska Academy at Sahlgrenska University Hospital, Göteborg, Sweden Received 16 December 2010
Abstract
Background: Increased beat-to-beat variability in cardiac repolarization time is a tentative risk marker of drug-induced torsades de pointes. We developed a new, automatic method based on the temporal variability of the T-wave down slope to assess this variability. Method and Results: Leads V1 to V6 of resting electrocardiograms were recorded in 42 healthy subjects (18-68 years, 22 men). The temporal variability at 50% of the T-wave down slope, deltaT50 (1.5 ± 0.41 milliseconds; range, 0.86-2.66 milliseconds), was measured with an accuracy of 1 millisecond on at least 9 pairs of electrocardiogram complexes with a signal-to-noise ratio more than 10 and changes in the R-R interval less than 150 milliseconds. The correlation between repeated measurements of deltaT50 was high. DeltaT50 was measured without corrections for age, sex, heart rate, T-wave amplitude, signal-to-noise ratio, R-R variability, and QTcF because none of these factors explained more than 4% of the within-subject deltaT50 variability. Conclusion: The beat-to-beat repolarization variability was measured with high fidelity with the deltaT50 method and was a robust measure in healthy volunteers. © 2011 Elsevier Inc. All rights reserved.
Keywords:
Beat-to-beat; QT variability; Repolarization instability; Automatic analysis
Introduction Variability in the cardiac repolarization time may indicate an electrical instability that constitutes a potential risk of lifethreatening arrhythmias.1 Both increased variability and prolongation of the cardiac repolarization time are considered to be indicators of a reduced repolarization reserve, which may imply an increased risk of torsades de pointes tachycardia (TdP).2,3 However, the 2 variables reflect different aspects of a reduced repolarization reserve because increased variability may occur at any QT interval duration,4 whereas drug-induced QT prolongation does not necessarily lead to an increased QT variability and TdP.5 In experimental studies, an increased beat-to-beat variability of the QT interval or of the monophasic action potential duration was superior to QT prolongation in predicting the risk of imminent TdP or sudden cardiac death.5-8 The potential value of QT variability as a diagnostic marker for TdP was ⁎ Corresponding author. AstraZeneca R&D, Mölndal and Sahlgrenska Academy at Sahlgrenska University Hospital, 431 83 Mölndal, Sweden. E-mail address: [email protected] 0022-0736/$ – see front matter © 2011 Elsevier Inc. All rights reserved. doi:10.1016/j.jelectrocard.2011.04.010
demonstrated in patients with congenital long-QT syndrome (LQTS), where combined measurements of the QT variability and the QT interval improved the identification of mutation carriers as compared with QT interval measurements alone.9 Hence, increased beat-to-beat repolarization variability may be a valuable risk marker for drug-induced proarrhythmia and a potential tool for identifying patients at increased risk of such proarrhythmia, but would require a precise and accurate method that automatically determines the cardiac repolarization time on a beat-to-beat basis. Methods that are available for assessing beat-to-beat repolarization variability include algorithms such as the tangent method,10 template matching,11 or measurement to the maximum derivative of the T-wave down slope.12 Initial work to measure beat-to-beat QT variability using the tangent method in EClysis, a computerized system for electrocardiogram (ECG) analysis, 10 showed that this algorithm worked appropriately in ECG recordings that were of good quality and had a normal QRST morphology, whereas a more robust method seemed to be needed to enable analysis of less perfect ECG signals on a single-beat basis. The idea of measuring the repolarization variability along the T-wave down slope
477.e2
C. Abrahamsson et al. / Journal of Electrocardiology 44 (2011) 477.e1–477.e9
emerged after superimposing a large number of consecutive ECG complexes from one ECG lead, as the T-wave down slope appeared to shift in parallel when the QT interval varied. Thus, to enable automatic measurement of the QT interval variability, a novel algorithm, deltaT50, which measures the temporal variability of the steep part of the T-wave down slope instead of the end of the T wave, was developed. Furthermore, to allow for automatic analysis of a large number of ECG complexes, the new algorithm was incorporated into a special version of EClysis used for research purposes. This study describes the rationale and development of the deltaT50 method in young, healthy subjects and presents initial results of application of the method in studies in healthy men and women of different ages.
Methods All subjects gave their written informed consent to participate in the study. The study protocols were approved by the ethics committee of Gothenburg University. ECG recording and definitions of measured variables Precordial ECG leads V1 to V6 were recorded during 10 minutes of supine rest in the morning after a 1 night fast. The ECG signals were sampled at 500 Hz and imported to EClysis.10 The EClysis system includes software that processes digitally recorded ECG signals from up to 12 leads at 500 Hz, using strictly defined algorithms to detect the PQRSTU points and to measure ECG intervals and amplitudes. The interday and intraday variability in major ECG intervals and amplitudes in healthy men and women was determined.13 The high-precision ECG measurements were validated by showing a high level of agreement when tested against reference values from a variety of ECGs from patients in the Common Standards for Quantitative Electrocardiography database.14 Hence, the deltaT50 algorithm was added to this validated version of EClysis, and the following calculations were done: on each single ECG complex, the intervals from the intersection between the RS line and the isoelectric line to 20%, 50%, and 80% of the T-wave down slope were defined as the T20, T50, and T80 intervals, respectively (Fig. 1A, C). If the RS line did not cross the isoelectric line, as in Fig. 1A, the interval started from the intersection between the extrapolated RS line (ie, the tangent of the RS line) and the isoelectric line. The averages of the absolute values of the beat-to-beat changes in the T20, T50, and T80 intervals were denoted deltaT20, deltaT50, and deltaT80, respectively (Fig. 1D). Accordingly, the average beat-to-beat change in the R-R intervals was denoted deltaRR (Fig. 1D). To reduce noise artifacts, the amplitude of each sample of the T-wave down slope was recalculated from the moving average of the amplitude of 21 consecutive samples, 10 before and 10 after the current sample (Fig. 1B).15 The signal-to-noise ratio (SNR) was calculated for each beat as described in the following. The signal amplitude was derived from the average T-wave amplitude during the interval between 50 milliseconds after the J point and the end of the
T wave. The noise amplitude was calculated as the average absolute difference in amplitude between the smoothed and the original curve in the diastolic interval from the end of the T wave to the start of the P wave. The SNR was subsequently calculated by dividing the signal amplitude by the noise amplitude. Following this procedure, the deltaT50 measurements were performed on a strict beat-to-beat basis. The deltaT50 method Using ECG recordings from 10 healthy subjects (6 men) aged 21 to 29 years, the deltaT50 method was developed and validated on the basis of the following 4 steps: (1) Measurements of deltaT20, deltaT50, and deltaT80 before and after smoothing. DeltaT20, deltaT50, and deltaT80 were calculated before and after smoothing on the first 30 beats each minute of each 10-minute ECG recording, that is, on 290 beat pairs/subject. (2) Investigation of the relationship observed between beat-to-beat changes in the T50 and R-R intervals. The correlation between the beat-to-beat changes in the T50 and the preceding R-R intervals was investigated on the first 30 beats of each minute of the 10-minute ECG recording in each subject, that is, on 290 beat pairs/subject, by applying linear regression to the data. (3) Effects of added noise to the ECG signals on deltaT20, deltaT50, and deltaT80. The first 30 beats of each ECG recording were analyzed to examine the influence of added noise and set criteria for acceptable signals. Beat pairs where the preceding R-R interval changed more than 150 milliseconds were excluded, which resulted in 27 to 29 beat pairs/ subject. Random white or 50-Hz sinusoidal noise of increasing amplitude (10, 50, 100, and 200 μV) was added to the ECG signals by means of a function in EClysis. The average deltaT20, deltaT50, and deltaT80 were calculated at each noise level before and after 21 p smoothing. The SNR was calculated before the smoothing. (4) Investigation of the least number of beats and the acceptable SNR for an estimation of deltaT50 with a precision of ±1 millisecond. To define the criteria for acceptable signals, that is, the least number of beat pairs and the acceptable noise level for an estimation of deltaT50 with a precision of ± 1 millisecond, a systematic investigation was undertaken in each subject. DeltaT50 was calculated at each noise level on a reduced number of beats (beats 1-5, 1-10, and 1-20) and compared with the deltaT50, resulting from the analysis of all 30 beats in the absence of noise.
DeltaT50 in healthy subjects at rest In addition to the 10 subjects who participated in the first part of the study described above, another 32 healthy subjects (16 men) aged 18 to 68 years were included to
C. Abrahamsson et al. / Journal of Electrocardiology 44 (2011) 477.e1–477.e9
477.e3
Fig. 1. A, An example of an ECG recording from one of the subjects in the study, illustrating the preceding R-R intervals and T50 intervals for a beat pair, beats 1 and 2. The R-R interval was the time between the R peaks, and the T50 interval was the time from the intersection between the RS line and the isoelectric line to 50% of the T-wave down slope (C). B, The original curve (black) was smoothed with 21 points (smoothed curve in gray) to reduce noise artifacts; that is, the amplitude of each sample was recalculated from the average amplitude of the preceding and following 10 samples together with the current sample. As an example, sample number 50 of the smoothed curve, marked with the curved arrow, is the average of samples 40 to 60 of the original curve, marked with the bracket. C, The time from the intersection of the RS line and the isoelectric line to 20%, 50%, and 80% of the T-wave down slope, the T20, T50, and T80 intervals, was measured on the ECG signals before and after smoothing. Interpolation between the discrete samples was used to determine the respective interval. C, The different levels marked on the smoothed curve, which is also shown in panel B. D, The definition of deltaT50 and deltaRR; that is, deltaT50 was the average change in the T50 intervals of consecutive beat pairs, and deltaRR was the average change in the preceding R-R intervals of consecutive beat pairs.
investigate the role of age, sex, SNR, T-wave amplitude, QTcF, heart rate (HR), and R-R variability (measured as deltaRR) for the short- and long-term variability of deltaT50. DeltaT50 was measured once every minute on 30 beats per 10 minutes on 2 occasions in each subject. In the first 10 subjects, the ECGs were recorded on 2 different days at an interval of 1 to 3 weeks; in the subsequent 32 subjects, ECGs were recorded on the same day but separated by 1 hour. The first 30 beats every minute were used for the calculation of deltaT50, and beat pairs where the preceding R-R interval changed more than 150 milliseconds or the SNR was less than 10 were excluded. This resulted in 9 to 29 beat pairs for the calculation of deltaT50. DeltaT50 was measured on both leads V2 and V5 in the 10 subjects and on either lead V5 or lead V2 in the 32 subjects. The HR and deltaRR were measured on all 30 beats. QTcF was automatically measured
on the average ECG complex of all 30 beats with the tangent method in EClysis.10
Statistical methods Statistical analyses were primarily performed using repeated-measurements analysis of variance, accounting for correlated observations for the subjects within and between the 2 study sessions. Analyses addressing subject factors, being based on the mean levels for each subject, for example, evaluating the potential relationship between deltaT50 and other variables of interest, were based on analysis of variance or linear regression, depending on whether only continuous or also categorical variables were included. P values less than .05 were regarded as statistically significant.
477.e4
C. Abrahamsson et al. / Journal of Electrocardiology 44 (2011) 477.e1–477.e9
Results The deltaT50 method (1) Measurements of deltaT20, deltaT50, and deltaT80 before and after smoothing. The effect of 21 p smoothing on deltaT20, deltaT50, and deltaT80 calculated on the first 30 beats each minute of the 10minute ECG recordings in the 10 subjects is depicted in Fig. 2. Smoothing decreased deltaT20, deltaT50, and deltaT80 significantly (P b .0001), but as seen in Fig. 2, the decrease was very small. The average deltaT20, deltaT50, and deltaT80 in the 10 subjects were 1.63 ± 0.69, 1.63 ± 0.71, and 1.84 ± 0.71 milliseconds before smoothing and 1.56 ± 0.69, 1.57 ± 0.70, and 1.62 ± 0.69 milliseconds after 21 p smoothing (Fig. 3A). (2) Investigation of the observed relationship between beat-to-beat changes in the T50 and R-R intervals. There was a positive and linear correlation between the beat-to-beat changes in the T50 and the R-R intervals in all subjects, and the average slope of the regression lines was 0.0178 (range, 0.0112-0.0247). However, the range of the beat-to-beat changes in the R-R interval varied substantially between subjects, between ±100 and ±500 milliseconds, as illustrated in Fig. 3B. Hence, to compare deltaT50 at a similar range of R-R interval changes in all subjects, we calculated deltaT20, deltaT50, and deltaT80 only on beat pairs with a change in the preceding R-R interval of less than 150 milliseconds (marked with b in Fig. 3B). Subsequently, when the relationship between deltaT50 and deltaRR was investigated, only 19% of
the variability in deltaT50 could be explained by variations in deltaRR (Fig. 6D; Table 1), and we decided not to correct deltaT50 for differences in deltaRR. With this criterion, deltaT20, deltaT50, and deltaT80 after smoothing were 1.30 ± 0.49, 1.28 ± 0.42, and 1.34 ± 0.36 milliseconds, respectively (Fig. 3A), and the between- and within-subject variability in deltaT50 decreased (Fig. 3C). (3) Effects of added noise to the ECG signals on deltaT20, deltaT50, and deltaT80. The original ECG signals were generally of good quality, and the average SNR was 87 (range, 40-169). After the addition of 10, 50, 100, and 200 μV of random white or 50-Hz sinusoidal noise, the average SNR decreased to 64, 21, 11, and 8 at the respective noise level. Because of the noise artifacts, deltaT20, deltaT50, and deltaT80 increased with decreasing SNR. Before smoothing, the increase became statistically significant when the average SNR was 21 or less (Fig. 4). DeltaT50 increased less than deltaT20 and deltaT80, implying that deltaT50 was the variable least sensitive to noise. 21 p smoothing reduced the noise artifacts, restored the morphology of the ECG signals (Fig. 4), and abolished the increase in deltaT20, deltaT50, and deltaT80 at an average SNR of 21 (Fig. 4). However, smoothing of the ECG signals was not sufficient to abolish the influence of noise on deltaT80 when the average SNR before smoothing was 11 or less, and on deltaT20 and deltaT50 when the average SNR was 6. Hence, deltaT80 was more sensitive to noise than deltaT20 and deltaT50. (4) Investigation of the least number of beats and the acceptable SNR for an estimation of deltaT50 with a precision of ±1 millisecond. The deltaT50 calculated on a reduced number of beats in the individual subjects showed that deltaT50 calculated on beats 1 to 10 (8 beat pairs in 1 subject and 9 beat pairs in the other 9 subjects) and 1 to 20 (17-19 beat pairs) never differed more than 1 millisecond in any subject from deltaT50 calculated on beats 1 to 30 (27-29 beat pairs) when the SNR was 10 or higher. Hence, the least number of beat pairs required was 9, and the acceptable SNR was 10 for an estimation of deltaT50 with a precision of ±1 millisecond. DeltaT50 in healthy subjects at rest
Fig. 2. DeltaT20, deltaT50, and deltaT80 were calculated on the first 30 beats each minute of the 10-minute ECG recording in the 10 subjects on the original curves (no smoothing) and after 21 points smoothing (for an explanation of the variables and the smoothing procedure, see Fig. 1). The correlation coefficients (r) for the relationship between deltaT20, deltaT50, and deltaT80 measured after smoothing and plotted as a function of the corresponding measures before smoothing are shown in the legend in the figure. The gray line represents the line of identity.
The average deltaT50 was 1.5 ± 0.41 milliseconds in the 42 subjects, ranging from 0.86 to 2.84 milliseconds (Fig. 5A), and deltaT50 measured on the 2 occasions in each subject correlated well (Fig. 5A). Furthermore, the variability of deltaT50 over the 10-minute period in each subject was small, and deltaT50 measured during the first minute of the ECG recording correlated well with the average deltaT50 measured over the entire 10-minute period (Fig. 5B). The results of the linear regression analysis of the correlation between deltaT50 and HR, T-wave amplitude, SNR, deltaRR, and QTcF, respectively, are summarized in
C. Abrahamsson et al. / Journal of Electrocardiology 44 (2011) 477.e1–477.e9
477.e5
Fig. 3. A, The average deltaT20, deltaT50, and deltaT80 in 10 young, healthy subjects calculated on the first 29 beat pairs each minute of a 10-minute ECG recording, before smoothing (bars to the left), after 21 points smoothing (bars in the middle, marked with a), and only on the beat pairs where the change in the preceding R-R interval was less than 150 milliseconds (bars to the right, marked with b). Mean ± SEM. B. The change in the T50 interval of the 290 beat pairs (T50n − T50(n − 1)) described in panel A vs the change in the preceding R-R interval (R-Rn − R-R(n − 1)) for the same beat pairs (1 symbol per subject). All beat pairs were used to calculate the average deltaT20, deltaT50, and deltaT80 shown in the bars in the middle, marked with a, in panel A. Only the beat pairs where the absolute change in the preceding R-R interval was less than 150 milliseconds were used to calculate the average deltaT20, deltaT50, and deltaT80 shown in the bars to the right, marked with b, in panel A. C, Individual values of deltaT50 in the 10 young subjects each minute of a 10-minute ECG recording (time in minutes on the x-axis). The beats marked with a in panel C were used to calculate deltaT50 in the left graph, and the beats marked with b in panel C were used to calculate deltaT50 in the right graph.
Table 1. Within subjects, deltaT50 correlated significantly with all variables except QTcF, but the R2 values were low (range, 0.5-3.5%), and the influence of each variable on the variability of deltaT50 was negligible (Table 1). Between
Table 1 Linear regression was used to investigate the correlation between deltaT50 and HR, T-wave amplitude, SNR, deltaRR, and QTcF within and between the 42 subjects during 10 minutes of rest DeltaT50 vs:
HR T-wave amplitude SNR DeltaRR QTcF
Within subjects
subjects, deltaT50 correlated significantly only with HR and deltaRR (Fig. 6A-E), but again, the R2 values were low, and variations in HR and deltaRR explained only 10% and 19%, respectively, of the total deltaT50 variability (Table 1). As a comparison, 47% and 52% of the within- and betweensubject variability in the QT interval, respectively, was explained by variations in the HR. Finally, there was no correlation between deltaT50 and age or sex, and deltaT50 measured from lead V2 differed less than 1 millisecond from deltaT50 measured from lead V5 (Fig. 7).
Between subjects
P
R2 (%)
P
R2 (%)
Discussion
b.0001 .018 .041 b.0001 NS
2.6 0.7 0.5 3.5
.044 NS NS .004 NS
10
The parallel nature of the shift of the T-wave down slope between consecutive beats, which is visible when consecutive ECG complexes are superimposed, was demonstrated by almost identical deltaT20, deltaT50, and deltaT80 values. This fact, together with the notion that both deltaT20 and deltaT80 were marginally more sensitive to the effects of
19
The P and the R2 values for the respective correlations are shown. NS indicate nonsignificant.
477.e6
C. Abrahamsson et al. / Journal of Electrocardiology 44 (2011) 477.e1–477.e9
Fig. 4. The average deltaT20, deltaT50, and deltaT80 calculated on 27 to 29 beat pairs in the 10 subjects before and after smoothing and before and after added noise (random or sinusoidal noise). Only beat pairs with a change in the preceding R-R interval of less than 150 milliseconds were included. Values are expressed as mean ± SEM. Inserted are also representative recordings of the T wave from 30 beats from 1 subject, showing the effects of smoothing on the signal quality.
Fig. 5. A, The average deltaT50 for 10 minutes was measured on 2 occasions separated by 1 to 3 weeks in 10 subjects and by 1 hour in 32 subjects (R2 = 52%). B, DeltaT50 was stable for 10 minutes, and deltaT50 measured the first of the 10 minutes correlated well with the average deltaT50 of the 10 minutes (R2 = 75%).
added noise to the ECG signals, made us choose deltaT50 as the primary measure of the beat-to-beat variability in cardiac repolarization time. Compared with the presently used measures, which define the QT interval as the interval between the start of the QRS complex to the end of the T wave, the fiducial points of the T50 interval are more easily defined. The start of the T50 interval is set to the crossing point between the RS line and the isoelectric line and the end to 50% of the T-wave down slope. Thereby, variations in the width of the QRS complex are largely avoided as well as problems with determination of the end of the T wave. Initially, all beats were included in the calculation of the deltaT50 for a series of 30 beats. Not surprisingly, we found that a large alteration in the R-R interval was associated with a major change in the QT interval of the following beat. Pronounced R-R and QT variability was found in some, but not all, subjects, and the analysis was refined to exclude beats after a change in the preceding R-R interval of more than 150 milliseconds to calculate delta T50 at a similar range of R-R interval changes in all subjects. Noise artifacts usually disturb the analysis of single beats, but the 21 p smoothing, which had a negligible effect on deltaT50 measured on “clean” signals, efficiently removed the noise artifacts on ECG signals with an SNR above 10 and the error of the estimate of deltaT50 caused by these introduced artifacts. Beats with an SNR of less than 10 were therefore excluded from the calculation of deltaT50. With these exclusion criteria, 9 beat pairs were found to be sufficient to estimate deltaT50 with an accuracy of 1 millisecond; that is, values obtained from 9 pairs never differed by more than 1 millisecond as compared with values derived from 29 pairs in any individual subject. DeltaT50 was generally very low, on average, approximately 1.5 milliseconds. Studies of beat-to-beat QT variability using comparable or similar methods are scarce in the
C. Abrahamsson et al. / Journal of Electrocardiology 44 (2011) 477.e1–477.e9
477.e7
Fig. 6. The correlations between deltaT50 and HR (A), T-wave amplitude (B), SNR (C), deltaRR (D), and QTcF (E). The P and R2 values for the respective correlation are summarized in Table 1.
literature (summarized in Pueyo et al16). The root mean square of differences between consecutive QT intervals (measured from Q start to T end), that is, similar to the deltaT50 in the present study, was automatically analyzed from several hours of ambulatory ECGs in healthy subjects and patients with LQTS.17 In that study, the average root mean square of differences between consecutive QT intervals was approximately 7 milliseconds in the healthy subjects and
Fig. 7. DeltaT50 was measured on leads V2 and V5 in 10 subjects once every minute during 10 minutes of rest (10 values per subject, 1 symbol per subject). The dotted line represents the line of identity. The difference in deltaT50 measured on the 2 leads was generally very small (b1 millisecond) but consistently higher or lower at all time points in the same subject.
15 milliseconds in the LQTS patients.17 Hinterseer et al4,9,18 manually measured 30 consecutive QT intervals from ECG printouts, and the short time variability, defined as the average beat-to-beat change in the QT interval divided by the square root of 2, was approximately 4 milliseconds in their control groups. Couderc et al19 used a wavelet transformation methodology and calculated the beat-to-beat shift of the resulting waveforms constituting the T wave and found a repolarization variability of 2 ± 4 milliseconds in healthy subjects. The deltaT50 variability over time was low, and the reproducibility of the measurement when carried out 1 hour or 1 to 3 weeks apart in the same individual was good. Moreover, the deltaT50 calculated on the first minute of the 10-minute continuous ECG recording correlated well with the average deltaT50 for the entire recording period in all subjects. Consequently, the individual deltaT50 may be determined with appropriate accuracy at a single time point and may be monitored minute by minute. By excluding beats after pronounced changes in the R-R interval, only 19% of the total variability in deltaT50 between the subjects and only 3.5% within the subjects could be explained by the R-R variability. There was thus no cause for correcting deltaT50 for deltaRR. The lack of correlation between deltaT50 and QTcF emphasizes that the 2 measures of cardiac repolarization ability are unconnected. In fact, deltaT50 was measured without correction for any of the other tentatively confounding factors,20 HR, T-wave amplitude, and SNR, that varied between the subjects or within the same subject over time. Considering that measurements of the QT interval involve rate correction formulae that overcorrect and undercorrect values deviating
477.e8
C. Abrahamsson et al. / Journal of Electrocardiology 44 (2011) 477.e1–477.e9
from the HR of 60 beats/min, this is a great advantage. The choice of ECG lead (V2 or V5) was also found to be of minor importance as long as the measurements over time were based on recordings from the same lead. A risk marker for proarrhythmias should ideally have a very high specificity and be measured with high precision in an automatic and operator-independent manner. With the deltaT50 method, it is possible to determine the beat-to-beat repolarization variability automatically with high precision on only 9 beat pairs, and the method was found to be robust and required no correction for variations in factors such as age, sex, QTcF, T-wave amplitude, SNR, or time dependence. The stability of deltaT50 when followed over time enables detection of small deviations from the baseline value. The significance of beat-to-beat repolarization variability as a risk marker is under evaluation, and ECG databases, such as those in the Telemetric and Holter ECG Warehouse, 21 may provide opportunities to test new methods in ECGs of subjects, patients on various pharmacologic agents, and patients who have experienced spontaneous or drug-induced ventricular arrhythmias.
Limitations Excluding beats after large changes in the R-R interval was a way to standardize the measurements, and we cannot rule out that the variability in repolarization after the largest changes in the R-R interval may contain important information to be considered in future studies with deltaT50. For example, after treatment with ibutilide, the QT variability index increased only during random interval right atrial pacing and not during sinus rhythm, possibly implying that measurements at baseline alone may not always disclose prevalent abnormal phenomena.22 The relationship between beat-to-beat changes in the T50 and R-R intervals should therefore be characterized for any study population other than healthy volunteers. The primary objective of the present study was to describe the development and the validation of a novel methodology to assess temporal variability of ventricular repolarization. To this end, we applied the technique on digital ECGs recorded from healthy subjects studied under standardized conditions. However, for deltaT50 to become a clinically useful tool, it must also work reliably in patients with apparently normal or abnormal ECGs as a consequence of, for example, disease or medication. Evaluation of the methodology in such patient populations is currently ongoing.
Conclusion The beat-to-beat repolarization variability in healthy subjects was automatically measured with the deltaT50 method with an accuracy of 1 millisecond on at least 9 beat pairs with an SNR more than 10 and changes in the preceding R-R interval less than150 milliseconds. With these criteria, the variability of deltaT50 over time was low, and
correction for differences in age, sex, HR, T-wave amplitude, SNR, deltaRR, and QTcF was not needed. Acknowledgments We are indebted to Lars Lind, MD, PhD, and Catherine Åhlund, RN, for their contribution to the studies from which data were collected. We are also grateful to John Gustafsson, PhD, for fruitful input and discussions during the initial development of the method. References 1. Berger RD. QT variability. J Electrocardiol 2003;36(Suppl):83. 2. Thomsen MB, Matz J, Volders PGA, Vos MA. Assessing the proarrhythmic potential of drugs: current status of models and surrogate parameters of torsades de pointes arrhythmias. Pharmacol Therapeut 2006;112:150. 3. Roden DM. Long QT syndrome: reduced repolarization reserve and the genetic link. J Int Medicine 2006;259:59. 4. Hinterseer M, Thomsen MB, Beckman B-M, et al. Beat-to-beat variability of QT intervals is increased in patients with drug-induced long-QT syndrome: a case control pilot study. Eur Heart J 2008;29:185. 5. Carlsson L, Andersson B, Linhardt G, Löfberg L. Assessment of the ion channel-blocking profile of the novel combined ion channel blocker AZD1305 and its proarrhythmic potential versus dofetilide in the methoxamine-sensitized rabbit in vivo. J Cardiovasc Pharmacol 2009;54:82. 6. Thomsen MB, Verduyn SC, Stengl M, et al. Increased short-term variability of repolarization predicts D-sotalol–induced torsades de pointes in dogs. Circulation 2004;110:2453. 7. Thomsen MB, Truin M, van Opstal JM, et al. Sudden cardiac death in dogs with remodelled hearts is associated with larger beat-to-beat variability of repolarization. Basic Res Cardiol 2005;100:279. 8. Thomsen MB, Volders PGA, Beekman JDM, Matz J, Vos MA. Beat-tobeat variability of repolarization determines proarrhythmic outcome in dogs susceptible to drug-induced torsades de pointes. J Am Coll Cardiol 2006;48:1268. 9. Hinterseer M, Beckman B-M, Thomsen MB, et al. Relation of increased short-term variability of QT interval to congenital long-QT syndrome. Am J Cardiol 2009;103:1244. 10. Dota C, Skallefell B, Edvardsson N, Fager G. Computer-based analysis of dynamic QT changes: toward high precision and individual rate correction. Ann Noninvasive Electrocardiol 2002;7:289. 11. Berger RD, Kasper EK, Baughman KL, Marban E, Calkins H, Tomaselli GF. Beat-to-beat QT interval variability. Novel evidence for repolarization lability in ischemic and nonischemic dilated cardiomyopathy. Circulation 1997;96:1557. 12. Mortara DW. Automated QT measurement and application to detection of moxifloxacin-induced changes. Ann Noninvasive Electrocardiol 2008;14(Suppl 1):S30. 13. Dota CD, Edvardsson N, Skallefell B, Fager G. PC-based ECG waveform recognition-validation of novel software against a reference ECG database. Ann Noninvasive Electrocardiol 2009;14(Suppl 1):S42. 14. Dota CD, Edvardsson N, Schützer KM, et al. Inter- and intraday variability in major electrocardiogram intervals and amplitudes in healthy men and women. PACE 2003;26:361. 15. Bracewell R. Noise reduction by smoothing. In: Bracewell R, editor. Fourier analysis and imaging. New York: Kluwer Academic/Plenum Publishers; 2003. p. 269. 16. Pueyo E, Martinez JP, Laguna P. Cardiac repolarization analysis using the surface electrocardiogram. Phil Trans R Soc A 2009;367:213. 17. Němec J, Buncová M, Shusterman V, Winter B, Shen W-K, Ackerman MJ. QT interval variability and adaption to heart rate changes in patients with long QT syndrome. PACE 2009;32:72. 18. Hinterseer M, Beckmann B-M, Thomsen MB, et al. Usefulness of shortterm variability of QT intervals as a predictor for electrical remodelling and proarrhythmia in patients with nonischemic heart failure. Am J Cardiol 2010;106:216.
C. Abrahamsson et al. / Journal of Electrocardiology 44 (2011) 477.e1–477.e9 19. Couderc J-P, Zareba W, Burattini L, Moss AJ. Beat-to-beat repolarization variability in LQTS patients with the SCN5A sodium channel gene mutation. PACE 1999;22:1581. 20. Couderc J-P. Cardiac regulation and electrocardiographic factors contributing to the measurement of repolarization variability. J Electrocardiol 2009;42:494.
477.e9
21. Couderc J-P, Zareba W. An update on QT measurements and interpretation methodologies. Ann Noninvasive Electrocardiol 2008;14 (Suppl 1):S1. 22. Cheng A, Dalal D, Fetics BJ, et al. Ibutilide-induced changes in temporal lability of ventricular repolarization in patients with and without structural heart disease. J Cardiovascular Electrophysiol 2009;8:873.