- Email: [email protected]
Post infarction risk stratification using the 3-D angle between QRS complex and T-wave vectors
Journal of Electrocardiology Vol. 37 Supplement 2004
Post Infarction Risk Stratification Using the 3-D Angle Between QRS Complex and T-Wave Vectors Marek Malik, PhD, MD, Katerina Hnatkova, PhD, and Velislav N. Batchvarov, MD,
Abstract: Present experience with prospective identification of patients who might benefit from prophylactic antiarrhythmic intervention is restricted to risk stratification using left ventricular ejection fraction (LVEF). The precision of LVEF-based identification of high risk patients is neither highly sensitive nor highly specific. This study investigated risk stratification of 466 survivors of acute myocardial infarction (86 women, mean age 57.5 years) for whom a 5-year follow-up was available. During the follow-up 67 patients died and 24 of these events were sudden arrhythmic deaths. In addition to LVEF, patients were stratified by mean heart rate, heart rate variability and the slope of heart rate turbulence, all derived from 24-hour Holter recording obtained before hospital discharge, and by the 3D angle between QRS complex and T wave vectors (total cosine R-to-T, TCRT) obtained from digital resting electrocardiogram before hospital discharge. Individual risk characteristics and their combinations were evaluated by calculating the areas under the receiver operator characteristics (ROC). The bootstrap technology was used to investigate these statistically. For the stratification of both all cause mortality and sudden arrhythmic death, TCRT was the strongest risk stratifier (area under ROC of 0.6857 ⫾ 0.0367, and 0.7275 ⫾ 0.0544, respectively) that compared very favourably to LVEF (area under the ROC of 0.6610 ⫾ 0.0362 and 0.6346 ⫾ 0.0595, for all cause and arrhythmic death prediction, both P ⬍ 10⫺10 for the comparison with TCRT). TCRT was also stronger in combination with other stratifiers, eg, TCRT ⫹ LVEF (area under ROC of 0.7631 ⫾ 0.0325 and 0.8057 ⫾ 0.0473, for all cause and arrhythmic death prediction) was stronger than mean heart rate ⫹ LVEF (area under ROC of 0.7396 ⫾ 0.0298 and 0.7673 ⫾ 0.0445, respectively, both P ⬍ 10⫺10 for comparison with TCRT ⫹ LVEF). Hence the 3D QRS-T angle is a very powerful risk stratifier especially suited for the prediction of sudden arrhythmic death. It should be prospectively investigated in future trials of prophylactic antiarrhythmic interventions. Key words: Myocardial infraction, risk stratification antiarrhythmic intervention, QRS-T angle.
From the Division of Cardiac and Vascular Sciences, St George’s Hospital Medical School, London, England. Reprint requests: Marek Malik, PhD, MD, Department of Cardiac and Vascular Sciences, St George’s Hospital Medical School, Cranmer Terrace, London SW17 0RE; e-mail: [email protected] © 2004 Elsevier Inc. All rights reserved. 0022-0736/04/370S-0057$30.00/0 doi:10.1016/j.jelectrocard.2004.08.058
201
202
Journal of Electrocardiology Vol. 37 Supplement 2004
The recent advances in the prophylactic use of implantable cardioverter defibrillators (ICD) make it practical to offer prophylactic intervention to patients at particular risks of arrhythmic events. Nevertheless, the present experience is primarily restricted only to selecting patients with reduced left ventricular ejection fraction (LVEF) (1). While low values of LVEF identify patients who are at greater risk, the identification of patients at risk using LVEF is neither particularly sensitive nor specific. The precision of the risk assessment used in the recent ICD trials was rather poor (2,3) and the adoption of these strategies in clinical practice leads to potentially questionable health care expenditure. Selection of patients for prophylactic ICD implantation is therefore frequently restricted by other criteria (eg, prolonged QRS duration) that have not been prospectively tested and that are introduced merely for restrictive economic reasons. In the past, a variety of additional risk stratifiers have been proposed. Unfortunately, the practicality of some of the risk characteristics is limited. For instance, several risk stratifiers in survivors of acute myocardial infarction (MI) are derived from 24hour Holter recordings but the acquisition of such a recording may require prolonging the in-hospital stay by one extra day. Other simple risk stratifiers have not fulfilled the expectations (eg, QT dispersion) or are used only infrequently (eg, detection of late potentials). For all these reasons, the research of novel risk stratifiers is still warranted. Identification of patients who might benefit from prophylactic antiarrhythmic interventions is also well known to suffer from the paradox that the more accurate the identification of patients at risk, the lower the sensitivity (4). In particular, there is presently no established technology that might be proposed for arrhythmic risk stratification of general public. While it is plausible to speculate that the assessment of LVEF and/or Holter based risk characteristics such as heart rate variability (HRV) might identify patients with subclinical heart disease in whom sudden cardiac death might be the first manifestation of the pathology, the technical requirements of such a screening are not feasible. Recently, it has been observed that subtle morphological characteristics of the standard 12-lead electrocardiogram (ECG) might also be used for risk assessment. The use of standard ECG is particular appealing because the acquisition of the usual 10second signal is simple, practical, and inexpensive. Among the numerical quantifiers derived from the standard 12-lead ECG, the angle between the QRS complex and T-wave vectors has been repeatedly studied. Increased vectorial deviation between
the spatial orientation of the QRS complex and of the T wave has been shown to identify patients at great risk of general post infarction complications (5), to lead to poor outcome in a general elderly population (6), as well as to increased mortality rate during long-term follow-up of patients with established ischemic heart disease (7). Nevertheless, it has never been properly investigated to which extent the increased QRS-T angle identifies patients at increased arrhythmic risk rather than at risk of all-cause mortality and/or other overall complications. Having this in mind, this study investigated the predictive power of the 3D QRS-T angle in a population of survivors of acute MI in whom long-term follow-up data were available together with several other recognised risk characteristics including LVEF and HRV.
Materials and Methods The study utilized the database and ECG collections of Post-Infarction Research Survey of St George’s Hospital in London. The details of the survey were previously published (8). In this study, the data of 466 patients (86 women, mean age 57.5 ⫾ 8.5, range 26 - 74 years at the index MI, 396 thrombolyzed, 234 treated with beta-blockers) were used. These patients were all those in the Survey for whom the complete following data were available: 1. LVEF based either on radionuclide scan assessment or on angiography, 2. Fully analysable 3-lead 24-hour Holter recording obtained prior to hospital discharge, 3. Digital orthogonal 3-lead XYZ ECG sampled at 1 kHz obtained also before hospital discharge. As previously described, the 24-hour Holter recording was used to derive the overall HRV value expressed by HRV index (HRVi), mean sinus rhythm heart rate measured in beats per minute, and slope of heart rate turbulence (TS) measured in milliseconds per RR interval. The Holter recordings were considered analysable if the sum of the durations of sinus rhythm RR intervals was ⱖ18 hours. If a Holter recording contained no isolated ectopic beats preventing the evaluation of heart rate turbulence, TS value was replaced with a constant that was sufficiently large to indicate no risk regardless of TS dichotomy. The digital orthogonal XYZ electrocardiograms were used to derive the so-called total cosine of the
TCRT and Post-Infarction Risk •
Malik, Hnatkova, and Batchvarov 203
Table 1. Numerical Comparison of Risk Stratifiers in Survivors and Non survivors All-cause Mortality Stratifier LVEF [%] Mean HR [bpm] HRVi [u] TS [ms/RR] TCRT
Sudden Arrhythmic Death
Survivors
Non survivors
P Value
Survivors
Non survivors
P Value
46.78 ⫾ 14.16 66.55 ⫾ 16.17 27.2 ⫾ 9.48 9.64 ⫾ 9.50 ⫺0.08 ⫾ 0.70
41.71 ⫾ 15.57 71.89 ⫾ 12.23 23.88 ⫾ 9.58 7.2 ⫾ 8.52 ⫺0.43 ⫾ 0.65
.005692 .000870 .007417 .007372 .000008
46.26 ⫾ 14.45 67.18 ⫾ 15.81 26.83 ⫾ 9.43 9.29 ⫾ 9.29 ⫺0.11 ⫾ 0.70
40.14 ⫾ 14.92 72.32 ⫾ 12.61 23.27 ⫾ 11.26 7.52 ⫾ 10.5 ⫺0.62 ⫾ 0.55
.057456 .039768 .063706 .139995 .000075
For each risk factor, the table shows mean values ⫾ standard deviation in survivors and non-survivors. The P values correspond to a non- parametric comparison by U-test. LVEF, left ventricular ejection fraction; HR, heart rate; HRVi, heart rate variability index; TS, slope of heart rate turbulence; TCRT, total cosine of the R to T angle.
R-to-T angle (TCRT) using previously published technology (9). For the purposes of the study, a 5-year follow-up was considered. During this period of follow-up, 67 patients died of all causes and of these, 24 patients suffered from sudden arrhythmic death. There are several possibilities of defining sudden cardiac death which differ in the relationship of the events to the probability of an arrhythmic cause (10). To ensure the tightest possible relationship between the arrhythmic classification and the arrhythmic cause, this study used the definition of sudden cardiac death as a death within one hour of the onset of new symptoms together with the absence of any indication (eg, from post mortem) that the death was not arrhythmic. In addition to the comparison of the numerical values of the characteristics between survivors and nonsurvivors, univariate and multivariate survival analysis (Cox proportional hazard model) was performed. For the purposes of these analyses, all the characteristics were dichotomized at the median value of the total population. While the statistical comparison of the different risk factors allows the numerical evaluation of their values and the comparison of simple median dichotomies, the practicality of different characteris-
tics is very strongly dependent on the dichotomies used to define the test positive cases. Moreover, it is well-known that optimum dichotomies of risk characteristics might strongly depend on the combination of risk factors since the optimum dichotomies for univariate risk assessment are frequently different from the optimum dichotomies in a multivariate setting (8). For all these reasons, the analytical core of this study used the comparison of ROC curves derived from individual risk factors and their combination (11). This made it possible to evaluate the predictive strength of the different stratifiers independent of any dichotomies. ROCs were calculated for all 5 individual risk factors considered in the study as well as for several pairs, triplets, and quadruplets of factors. A previously published technology was employed that combined variable dichotomies of combined risk stratifiers. The characteristics based on 2 variables acquired both of them to be positive, the characteristics based on triplets of variables required at least 2 of the 3 to be positive, and the characteristics based on quadruplets of the stratifiers required at least 3 of them to be positive (12). In other words, ROC calculated for the combination of 2 risk characteristics ⌽ and ⌿ varied their
Table 2. Univariate and Multivariate Association of Dichotomized Stratifiers With Follow-up Events All-cause Mortality
Sudden Arrhythmic Death
Stratifier
Univariate
Multivariate
Univariate
Multivariate
LVEF Mean HR HRVi TS TCRT
0.01053 0.00153 0.06077 0.29637 0.00028
0.19672 0.02674 0.57465 0.88535 0.00165
0.15223 0.01114 0.58334 0.34826 0.00052
0.69032 0.04035 0.65067 0.63658 0.00350
For each stratifier, the table shows p-values of statistical tests of association of the risk characteristic dichotomised at the median of the total population with follow-up outcome. The univariate results are the p-values of log-rank test investigating each variable separately, the multivariate results are p-values obtained with Cox proportional hazard model. Abbreviations as in Table 1
204 Journal of Electrocardiology Vol. 37 Supplement 2004
Fig. 1. Kaplan-Mayer survival curves of the all cause survival (top) and sudden arrhythmic death free survival (bottom) for the population stratified according to the median value of total cosine R-to-T (TCRT) and to the median value of left ventricular ejection fraction (LVEF). SCD ⫽ sudden cardiac death.
dichotomies and through all possible values and for each combination of and established the sensitivity and specificity of identification of high risk patients as those for whom (⌽ ⱕ and ⌿ ⱕ ).† From this process, the maximum achievable specificity for each level of sensitivity was obtained which defined the ROC curve as the dependency of specificity on sensitivity. Similarly, ROC calculated for the combination of three characteristics ⌽, ⌿ and ⍀ varied their dichotomies , and through all possible values, and for each set of dichotomies established the sensitivity and specificity of identification of high risk patients as those for whom 储⌽ ⱕ 储 ⫹ 储⌿ ⱕ 储 ⫹ 储⍀ ⱕ 储 ⱖ 2, where for a logical expression , the symbol 储储 is 1 or 0 if is or is not valid, respectively. A corresponding procedure was used for combinations of four characteristics. For each ROC, the area under the curve was calculated. To allow statistical evaluation of the areas under ROC curves, the bootstrap method was used to repeat the calculation of each ROC (13). The formula assumes that low values of both and ⌿ identify high risk. Opposite relations were used for risk factors with which high values identify high risk (eg, heart rate.) †
Specifically, the N patients of the population of the study were numbered {p(1),p(2),…,p(N)}, N ⫽ 466. Subsequently, for each considered combination of risk factors {⌶1,…,⌶w} where w ⫽ 1,…,4, the following bootstrap algorithm was repeated 1,000 times: 1. A sequence N of whole random numbers {1,2,3,…,N} between 1 and N (with repetitions) was generated and a new population of patients {p(1),p(2),p(3),…,p(N)} was considered, i.e. if x ⫽ y, patient p(x) was considered more than once, whilst if m ⰻ {1,2,3,…,N}, patient p(m) was not considered. 2. The ROC curve corresponding to characteristics {⌶1,…,⌶w} was calculated for the population {p(1),p(2),p(3),…,p(N)} considering the fact that different random selections led to a different number of patients with follow-up events. 3. The area under the calculated ROC curve was obtained. In this way, 1,000 samples of the area under the ROC curve were obtained for each considered combination of risk factors {⌶1,…,⌶w}. In addition to
TCRT and Post-Infarction Risk •
Malik, Hnatkova, and Batchvarov 205
the statistical summary of these values for each combination of risk factors, the repeated values of the area under the ROC curve were also compared between different combinations of risk factors using nonparametric statistical tests.
Results Table 1 shows the statistical comparison between the numerical values of different risk stratifiers between survivors and non-survivors distinguishing all-cause mortality and sudden cardiac death. As expected, all the considered characteristics led to statistical significant differences between the mean values in survivors and nonsurvivors of all-cause mortality. Table 2 shows the univariate association between different characteristics dichotomised at their medians with the incidence of all cause mortality and sudden cardiac death during the follow-up. It can be very clearly observed that of the 5 risk stratifiers, the 3D QRS-T angle was the strongest risk factor for both identification of the risk of all cause mortality and the risk of sudden cardiac death. Corresponding comparison of Kaplan-Meier survival curves for median dichotomy of LVEF and TCRT is shown in Figure 1. Figure 2 shows the values of the area under the ROC curves derived for both individual risk stratifiers and for their combinations. Corresponding numerical values are shown in Table 3. Both the table and figure show that TCRT was statistically significantly the strongest risk stratifier for both all cause mortality and sudden arrhythmic death. This was not only true in a univariate setting but also in combinations of different risk factors. Each time adding TCRT to a battery of stratifiers made a significant improvement in the identification of high risk cases. The effect of adding TRCT to the stratification strategies is even more clearly seen in Figure 3 that shows the cumulative results of the repeated bootstrap experiments. This figure also shows that while compared to LVEF, TCRT was a stronger predictor of death in this population, the improvement in predictive power of sudden arrhythmic death was much more substantial.
Discussion The study shows convincingly that pathological values of the 3D QRS-T angle are not only a very
Fig. 2. Areas under the curve (AUC) of receiver operator characteristics (ROC) obtained with different combinations of risk factors used to predict the cases of all cause mortality (top part) and of sudden arrhythmic death (bottom part). Open marks correspond to the risk factors listed at the horizontal axes, light full squares to the listed risk factors combined with left ventricular ejection fraction, and dark full squares to the listed risk factors combined with total cosine R-to-T. Abbreviations as in Table 1.
potent risk stratifier of post-infarction patients who are at increased risk of sudden arrhythmic death. While the predictive value of all the individual characteristics, including TCRT, is rather modest, the areas under ROCs obtained for com-
206 Journal of Electrocardiology Vol. 37 Supplement 2004 Table 3. Areas Under the Receiver Operator Characteristics Obtained for Risk Stratification All-cause mortality Fisk Factor(s)
Mean
95% Confidence Interval
LVEF TCRT LVEF ⫹ HR LVEF ⫹ HRVi LVEF ⫹ TS LVEF ⫹ TCRT LVEF ⫹ TCRT ⫹ HRVi LVEF ⫹ TCRT ⫹ TS LVEF ⫹ TCRT ⫹ HR HRVi ⫹ TCRT ⫹ TS HRVi ⫹ TCRT ⫹ HR LVEF ⫹ HRVi ⫹ TCRT ⫹ TS LVEF ⫹ HRVi ⫹ TCRT ⫹ HR
0.6610 0.6857 0.7396 0.7197 0.6984 0.7631 0.8067 0.8057 0.7988 0.7933 0.7995 0.8289 0.8359
(0.6588–0.6633) (0.6834–0.6880) (0.7377–0.7414) (0.7178–0.7215) (0.6965–0.7004) (0.7611–0.7651) (0.8049–0.8084) (0.8040–0.8073) (0.7971–0.8005) (0.7916–0.7951) (0.7978–0.8012) (0.8273–0.8304) (0.8344–0.8375)
Arrhythmic Death Mean
95% Confidence Interval
0.6346 0.7275 0.7673 0.7540 0.6902 0.8057 0.8623 0.8344 0.8456 0.8507 0.8707 0.8789 0.8938
(0.6309–0.6383) (0.7241–0.7309) (0.7645–0.7700) (0.7514–0.7566) (0.6869–0.6935) (0.8028–0.8086) (0.8602–0.8644) (0.8318–0.8370) (0.8434–0.8479) (0.8485–0.8528) (0.8687–0.8726) (0.8769–0.8809) (0.8920–0.8956)
The table lists areas under the receiver operator characteristics calculated for different combinations of risk factors used to predict the cases of death during follow-up. Abbreviations as in Table 1 All the risk factor combinations with non-overlapping 95% confidence interval of means of the areas were also statistically different in non-parametric comparisons.
binations of risk stratifiers suggest that a combination of different characteristics may improve the practicality of the identification of patients who might benefit from prophylactic intervention very considerably. The observation of the strong association of pathological values of TCRT with poor outcome is in a good agreement with previously published results (5–7). In this sense, the observations made in this study are not surprising. It is also not surprising that pathological values of TCRT are particularly associated with the risk of sudden arrhythmic death. The pathologies of ventricular re-
polarisation clearly play an important role in the genesis and pathophysiology of proarrhythmic substrates. On the contrary, the predictive value of HRV and of heart rate turbulence slope observed in this investigation was low. This is at odds with previously published observations (14 –16) and it is indeed contradicting even the data found in other sections of this Post-Infarction Survey (17,18). The particular selection of patients might have introduced some bias. Since the digital ECGs were restricted to the orthogonal XYZ leads, we were unable to evaluate
Fig. 3. Cumulative distribution of the areas under the receiver operator characteristics obtained during the repeated bootstrap experiments. Abbreviations as in Table 1. See the text for the details of the bootstrap technology.
TCRT and Post-Infarction Risk •
other morphological characteristics that are obtainable from the standard 12-lead ECG. In particular, we were unable to study the so-called T wave and QRS residua (19) that had previously been observed to be even stronger indicators of risk in ischemic heart disease (7). Although we used the bootstrap method to investigate the areas under the ROC curves statistically, all the bootstrap repetitions were taken from a moderately sized population of MI survivors with a rather low 5-year incidence of sudden arrhythmic death. At the same time, however, the difference between the LVEF-based and TCRT-based stratification of patients at risk of arrhythmic death is so striking that it might have been only little influenced by the composite of this particular population. Since we intended to use a population with a follow-up duration similar to the longevity of an ICD, the study population did not fully reflect the present trends in the therapy of acute MI. Although a high proportion of the patients were thrombolysed, the treatment with beta-blockers was lower compared to the present practice. It has been shown that the predictive power of most previously established risk stratifiers is much lower if not directly lost in patients actively treated with beta-blockers (20). Nevertheless, our most recent (unpublished, March 2004) experience with repolarization morphology suggests that TCRT is little influenced by beta-blocking therapy. In spite of these limitations, the study demonstrates the predictive value of the increased QRS-T angle convincingly. Because the predictive value of the QRS-T angle has now been consistently shown in several independent studies, and since this study shows that the pathological values of the angle are strongly associated with arrhythmic death, a sound recommendation can be made that in forthcoming trials of prophylactic ICD implantation, the 3D QRS-T angle is incorporated into the selection of patients to be treated, probably in combination with lower LVEF and possibly also with other risk characteristics.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
References 15. 1. Gregoratos G, Abrams J, Epstein AE, et al: ACC/ AHA/NASPE 2002 guideline update for implantation of cardiac pacemakers and antiarrhythmia devices: A report of the American College of Cardiology/American Heart Association Task Force on Practice Guidelines (ACC/AHA/NASPE Committee
16.
Malik, Hnatkova, and Batchvarov 207
on Pacemaker Implantation). www.acc.org/clinical/guidelines/pacemaker/pacemaker.pdf. 2002. Greenberg H, Case RB, Moss AJ, et al: MADIT-II Investigators: Analysis of mortality events in the Multicenter Automatic Defibrillator Implantation Trial (MADIT-II). J Am Coll Cardiol 43:1459, 2004 Nisam S, Henry S, Wilber DJ: Implementation of MADIT and MUSTT in clinical practice: Results of an international survey. Ann Noninvasive Electrocardiol 7:399, 2002 Myerburg RJ, Mitrani R, Interian A Jr, et al: Interpretation of outcomes of antiarrhythmic clinical trials: Design features and population impact. Circulation 97:1514, 1998 Zabel M, Acar B, Klingenheben T: Analysis of 12-Lead T-wave morphology for risk stratification after myocardial infarction. Circulation 102:1252, 2002 Kardys I, Kors JA, van der Meer IM, et al: Spatial QRS-T angle predicts cardiac death in a general population. Eur Heart J 24:1357, 2003 Zabel M, Malik M, Hnatkova K, et al: Analysis of T-wave morphology from the 12-lead electrocardiogram for prediction of long-term prognosis in male US veterans. Circulation 105:1066, 2002 Redwood SR, Odemuyiwa O, Hnatkova K, et al: Selection of dichotomy limits for multifactorial prediction of arrhythmic events and mortality in survivors of acute myocardial infarction. Eur Heart J 18:1278, 1997 Acar B, Yi G, Hnatkova K, et al: Spatial, temporal and wavefront direction characteristics of 12-lead T wave morphology. Med Biol Eng Comput 37:574, 1999 Torp-Pedersen C, Kober L, Elming H, et al: Classification of sudden and arrhythmic death. Pacing Clin Electrophysiol 20:2545, 1997 Hanley JA, McNeil BJ: The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology 143:29, 1982 Hnatkova K, Poloniecki J, Camm AJ, et al: Computation of multifactorial receiver operator and predictive accuracy characteristics. Comp Meth Prog Biomed 42:147, 1994 Efron B, Tibshirani RJ: An introduction to the bootstrap, in Monographs on Statistics and Applied Probability, vol 57, London, Chapman & Hall, 1993 Heart rate variability: standards of measurement, physiological interpretation and clinical use. Task Force of the European Society of Cardiology and the North American Society of Pacing and Electrophysiology. Circulation 93:1043, 1996. Schmidt G, Malik M, Barthel P, et al: Heart rate turbulence after ventricular premature beats as a predictor of mortality after acute myocardial infarction. Lancet 353:1390, 1999 Barthel P, Schneider R, Bauer A, et al: Risk stratification after acute myocardial infarction by heart rate turbulence. Circulation 108:1221, 2003
208 Journal of Electrocardiology Vol. 37 Supplement 2004 17. Odemuyiwa O, Malik M, Farrell T, et al: A comparison of the predictive characteristics of heart rate variability index and left ventricular ejection fraction for all-cause mortality, arrhythmic events and sudden death after acute myocardial infarction. Am J Cardiol 68:434, 1991 18. Hartikainen JEK, Malik M, Staunton A, et al: Distinction between arrhythmic and nonarrhythmic death after acute myocardial infarction based on heart rate variability, signal-averaged electrocardiogram, ventric-
ular arrhythmias and left ventricular ejection fraction. J Am Coll Cardiol 28:296, 1996 19. Malik M, Acar B, Yi G, et al: QT dispersion does not represent electrocardiographic interlead heterogeneity of ventricular repolarization. J Cardiovasc Electrophysiol 11:835, 2001 20. Huikuri H, Tapanainen JM, Lindgren K, et al: Prediction of Sudden Cardiac Death After Myocardial Infarction in the Beta-Blocking Era. J Am Coll Cardiol 42:652, 2003
Post Infarction Risk Stratification Using the 3-D Angle Between QRS Complex and T-Wave Vectors Marek Malik, PhD, MD, Katerina Hnatkova, PhD, and Velislav N. Batchvarov, MD,
Abstract: Present experience with prospective identification of patients who might benefit from prophylactic antiarrhythmic intervention is restricted to risk stratification using left ventricular ejection fraction (LVEF). The precision of LVEF-based identification of high risk patients is neither highly sensitive nor highly specific. This study investigated risk stratification of 466 survivors of acute myocardial infarction (86 women, mean age 57.5 years) for whom a 5-year follow-up was available. During the follow-up 67 patients died and 24 of these events were sudden arrhythmic deaths. In addition to LVEF, patients were stratified by mean heart rate, heart rate variability and the slope of heart rate turbulence, all derived from 24-hour Holter recording obtained before hospital discharge, and by the 3D angle between QRS complex and T wave vectors (total cosine R-to-T, TCRT) obtained from digital resting electrocardiogram before hospital discharge. Individual risk characteristics and their combinations were evaluated by calculating the areas under the receiver operator characteristics (ROC). The bootstrap technology was used to investigate these statistically. For the stratification of both all cause mortality and sudden arrhythmic death, TCRT was the strongest risk stratifier (area under ROC of 0.6857 ⫾ 0.0367, and 0.7275 ⫾ 0.0544, respectively) that compared very favourably to LVEF (area under the ROC of 0.6610 ⫾ 0.0362 and 0.6346 ⫾ 0.0595, for all cause and arrhythmic death prediction, both P ⬍ 10⫺10 for the comparison with TCRT). TCRT was also stronger in combination with other stratifiers, eg, TCRT ⫹ LVEF (area under ROC of 0.7631 ⫾ 0.0325 and 0.8057 ⫾ 0.0473, for all cause and arrhythmic death prediction) was stronger than mean heart rate ⫹ LVEF (area under ROC of 0.7396 ⫾ 0.0298 and 0.7673 ⫾ 0.0445, respectively, both P ⬍ 10⫺10 for comparison with TCRT ⫹ LVEF). Hence the 3D QRS-T angle is a very powerful risk stratifier especially suited for the prediction of sudden arrhythmic death. It should be prospectively investigated in future trials of prophylactic antiarrhythmic interventions. Key words: Myocardial infraction, risk stratification antiarrhythmic intervention, QRS-T angle.
From the Division of Cardiac and Vascular Sciences, St George’s Hospital Medical School, London, England. Reprint requests: Marek Malik, PhD, MD, Department of Cardiac and Vascular Sciences, St George’s Hospital Medical School, Cranmer Terrace, London SW17 0RE; e-mail: [email protected] © 2004 Elsevier Inc. All rights reserved. 0022-0736/04/370S-0057$30.00/0 doi:10.1016/j.jelectrocard.2004.08.058
201
202
Journal of Electrocardiology Vol. 37 Supplement 2004
The recent advances in the prophylactic use of implantable cardioverter defibrillators (ICD) make it practical to offer prophylactic intervention to patients at particular risks of arrhythmic events. Nevertheless, the present experience is primarily restricted only to selecting patients with reduced left ventricular ejection fraction (LVEF) (1). While low values of LVEF identify patients who are at greater risk, the identification of patients at risk using LVEF is neither particularly sensitive nor specific. The precision of the risk assessment used in the recent ICD trials was rather poor (2,3) and the adoption of these strategies in clinical practice leads to potentially questionable health care expenditure. Selection of patients for prophylactic ICD implantation is therefore frequently restricted by other criteria (eg, prolonged QRS duration) that have not been prospectively tested and that are introduced merely for restrictive economic reasons. In the past, a variety of additional risk stratifiers have been proposed. Unfortunately, the practicality of some of the risk characteristics is limited. For instance, several risk stratifiers in survivors of acute myocardial infarction (MI) are derived from 24hour Holter recordings but the acquisition of such a recording may require prolonging the in-hospital stay by one extra day. Other simple risk stratifiers have not fulfilled the expectations (eg, QT dispersion) or are used only infrequently (eg, detection of late potentials). For all these reasons, the research of novel risk stratifiers is still warranted. Identification of patients who might benefit from prophylactic antiarrhythmic interventions is also well known to suffer from the paradox that the more accurate the identification of patients at risk, the lower the sensitivity (4). In particular, there is presently no established technology that might be proposed for arrhythmic risk stratification of general public. While it is plausible to speculate that the assessment of LVEF and/or Holter based risk characteristics such as heart rate variability (HRV) might identify patients with subclinical heart disease in whom sudden cardiac death might be the first manifestation of the pathology, the technical requirements of such a screening are not feasible. Recently, it has been observed that subtle morphological characteristics of the standard 12-lead electrocardiogram (ECG) might also be used for risk assessment. The use of standard ECG is particular appealing because the acquisition of the usual 10second signal is simple, practical, and inexpensive. Among the numerical quantifiers derived from the standard 12-lead ECG, the angle between the QRS complex and T-wave vectors has been repeatedly studied. Increased vectorial deviation between
the spatial orientation of the QRS complex and of the T wave has been shown to identify patients at great risk of general post infarction complications (5), to lead to poor outcome in a general elderly population (6), as well as to increased mortality rate during long-term follow-up of patients with established ischemic heart disease (7). Nevertheless, it has never been properly investigated to which extent the increased QRS-T angle identifies patients at increased arrhythmic risk rather than at risk of all-cause mortality and/or other overall complications. Having this in mind, this study investigated the predictive power of the 3D QRS-T angle in a population of survivors of acute MI in whom long-term follow-up data were available together with several other recognised risk characteristics including LVEF and HRV.
Materials and Methods The study utilized the database and ECG collections of Post-Infarction Research Survey of St George’s Hospital in London. The details of the survey were previously published (8). In this study, the data of 466 patients (86 women, mean age 57.5 ⫾ 8.5, range 26 - 74 years at the index MI, 396 thrombolyzed, 234 treated with beta-blockers) were used. These patients were all those in the Survey for whom the complete following data were available: 1. LVEF based either on radionuclide scan assessment or on angiography, 2. Fully analysable 3-lead 24-hour Holter recording obtained prior to hospital discharge, 3. Digital orthogonal 3-lead XYZ ECG sampled at 1 kHz obtained also before hospital discharge. As previously described, the 24-hour Holter recording was used to derive the overall HRV value expressed by HRV index (HRVi), mean sinus rhythm heart rate measured in beats per minute, and slope of heart rate turbulence (TS) measured in milliseconds per RR interval. The Holter recordings were considered analysable if the sum of the durations of sinus rhythm RR intervals was ⱖ18 hours. If a Holter recording contained no isolated ectopic beats preventing the evaluation of heart rate turbulence, TS value was replaced with a constant that was sufficiently large to indicate no risk regardless of TS dichotomy. The digital orthogonal XYZ electrocardiograms were used to derive the so-called total cosine of the
TCRT and Post-Infarction Risk •
Malik, Hnatkova, and Batchvarov 203
Table 1. Numerical Comparison of Risk Stratifiers in Survivors and Non survivors All-cause Mortality Stratifier LVEF [%] Mean HR [bpm] HRVi [u] TS [ms/RR] TCRT
Sudden Arrhythmic Death
Survivors
Non survivors
P Value
Survivors
Non survivors
P Value
46.78 ⫾ 14.16 66.55 ⫾ 16.17 27.2 ⫾ 9.48 9.64 ⫾ 9.50 ⫺0.08 ⫾ 0.70
41.71 ⫾ 15.57 71.89 ⫾ 12.23 23.88 ⫾ 9.58 7.2 ⫾ 8.52 ⫺0.43 ⫾ 0.65
.005692 .000870 .007417 .007372 .000008
46.26 ⫾ 14.45 67.18 ⫾ 15.81 26.83 ⫾ 9.43 9.29 ⫾ 9.29 ⫺0.11 ⫾ 0.70
40.14 ⫾ 14.92 72.32 ⫾ 12.61 23.27 ⫾ 11.26 7.52 ⫾ 10.5 ⫺0.62 ⫾ 0.55
.057456 .039768 .063706 .139995 .000075
For each risk factor, the table shows mean values ⫾ standard deviation in survivors and non-survivors. The P values correspond to a non- parametric comparison by U-test. LVEF, left ventricular ejection fraction; HR, heart rate; HRVi, heart rate variability index; TS, slope of heart rate turbulence; TCRT, total cosine of the R to T angle.
R-to-T angle (TCRT) using previously published technology (9). For the purposes of the study, a 5-year follow-up was considered. During this period of follow-up, 67 patients died of all causes and of these, 24 patients suffered from sudden arrhythmic death. There are several possibilities of defining sudden cardiac death which differ in the relationship of the events to the probability of an arrhythmic cause (10). To ensure the tightest possible relationship between the arrhythmic classification and the arrhythmic cause, this study used the definition of sudden cardiac death as a death within one hour of the onset of new symptoms together with the absence of any indication (eg, from post mortem) that the death was not arrhythmic. In addition to the comparison of the numerical values of the characteristics between survivors and nonsurvivors, univariate and multivariate survival analysis (Cox proportional hazard model) was performed. For the purposes of these analyses, all the characteristics were dichotomized at the median value of the total population. While the statistical comparison of the different risk factors allows the numerical evaluation of their values and the comparison of simple median dichotomies, the practicality of different characteris-
tics is very strongly dependent on the dichotomies used to define the test positive cases. Moreover, it is well-known that optimum dichotomies of risk characteristics might strongly depend on the combination of risk factors since the optimum dichotomies for univariate risk assessment are frequently different from the optimum dichotomies in a multivariate setting (8). For all these reasons, the analytical core of this study used the comparison of ROC curves derived from individual risk factors and their combination (11). This made it possible to evaluate the predictive strength of the different stratifiers independent of any dichotomies. ROCs were calculated for all 5 individual risk factors considered in the study as well as for several pairs, triplets, and quadruplets of factors. A previously published technology was employed that combined variable dichotomies of combined risk stratifiers. The characteristics based on 2 variables acquired both of them to be positive, the characteristics based on triplets of variables required at least 2 of the 3 to be positive, and the characteristics based on quadruplets of the stratifiers required at least 3 of them to be positive (12). In other words, ROC calculated for the combination of 2 risk characteristics ⌽ and ⌿ varied their
Table 2. Univariate and Multivariate Association of Dichotomized Stratifiers With Follow-up Events All-cause Mortality
Sudden Arrhythmic Death
Stratifier
Univariate
Multivariate
Univariate
Multivariate
LVEF Mean HR HRVi TS TCRT
0.01053 0.00153 0.06077 0.29637 0.00028
0.19672 0.02674 0.57465 0.88535 0.00165
0.15223 0.01114 0.58334 0.34826 0.00052
0.69032 0.04035 0.65067 0.63658 0.00350
For each stratifier, the table shows p-values of statistical tests of association of the risk characteristic dichotomised at the median of the total population with follow-up outcome. The univariate results are the p-values of log-rank test investigating each variable separately, the multivariate results are p-values obtained with Cox proportional hazard model. Abbreviations as in Table 1
204 Journal of Electrocardiology Vol. 37 Supplement 2004
Fig. 1. Kaplan-Mayer survival curves of the all cause survival (top) and sudden arrhythmic death free survival (bottom) for the population stratified according to the median value of total cosine R-to-T (TCRT) and to the median value of left ventricular ejection fraction (LVEF). SCD ⫽ sudden cardiac death.
dichotomies and through all possible values and for each combination of and established the sensitivity and specificity of identification of high risk patients as those for whom (⌽ ⱕ and ⌿ ⱕ ).† From this process, the maximum achievable specificity for each level of sensitivity was obtained which defined the ROC curve as the dependency of specificity on sensitivity. Similarly, ROC calculated for the combination of three characteristics ⌽, ⌿ and ⍀ varied their dichotomies , and through all possible values, and for each set of dichotomies established the sensitivity and specificity of identification of high risk patients as those for whom 储⌽ ⱕ 储 ⫹ 储⌿ ⱕ 储 ⫹ 储⍀ ⱕ 储 ⱖ 2, where for a logical expression , the symbol 储储 is 1 or 0 if is or is not valid, respectively. A corresponding procedure was used for combinations of four characteristics. For each ROC, the area under the curve was calculated. To allow statistical evaluation of the areas under ROC curves, the bootstrap method was used to repeat the calculation of each ROC (13). The formula assumes that low values of both and ⌿ identify high risk. Opposite relations were used for risk factors with which high values identify high risk (eg, heart rate.) †
Specifically, the N patients of the population of the study were numbered {p(1),p(2),…,p(N)}, N ⫽ 466. Subsequently, for each considered combination of risk factors {⌶1,…,⌶w} where w ⫽ 1,…,4, the following bootstrap algorithm was repeated 1,000 times: 1. A sequence N of whole random numbers {1,2,3,…,N} between 1 and N (with repetitions) was generated and a new population of patients {p(1),p(2),p(3),…,p(N)} was considered, i.e. if x ⫽ y, patient p(x) was considered more than once, whilst if m ⰻ {1,2,3,…,N}, patient p(m) was not considered. 2. The ROC curve corresponding to characteristics {⌶1,…,⌶w} was calculated for the population {p(1),p(2),p(3),…,p(N)} considering the fact that different random selections led to a different number of patients with follow-up events. 3. The area under the calculated ROC curve was obtained. In this way, 1,000 samples of the area under the ROC curve were obtained for each considered combination of risk factors {⌶1,…,⌶w}. In addition to
TCRT and Post-Infarction Risk •
Malik, Hnatkova, and Batchvarov 205
the statistical summary of these values for each combination of risk factors, the repeated values of the area under the ROC curve were also compared between different combinations of risk factors using nonparametric statistical tests.
Results Table 1 shows the statistical comparison between the numerical values of different risk stratifiers between survivors and non-survivors distinguishing all-cause mortality and sudden cardiac death. As expected, all the considered characteristics led to statistical significant differences between the mean values in survivors and nonsurvivors of all-cause mortality. Table 2 shows the univariate association between different characteristics dichotomised at their medians with the incidence of all cause mortality and sudden cardiac death during the follow-up. It can be very clearly observed that of the 5 risk stratifiers, the 3D QRS-T angle was the strongest risk factor for both identification of the risk of all cause mortality and the risk of sudden cardiac death. Corresponding comparison of Kaplan-Meier survival curves for median dichotomy of LVEF and TCRT is shown in Figure 1. Figure 2 shows the values of the area under the ROC curves derived for both individual risk stratifiers and for their combinations. Corresponding numerical values are shown in Table 3. Both the table and figure show that TCRT was statistically significantly the strongest risk stratifier for both all cause mortality and sudden arrhythmic death. This was not only true in a univariate setting but also in combinations of different risk factors. Each time adding TCRT to a battery of stratifiers made a significant improvement in the identification of high risk cases. The effect of adding TRCT to the stratification strategies is even more clearly seen in Figure 3 that shows the cumulative results of the repeated bootstrap experiments. This figure also shows that while compared to LVEF, TCRT was a stronger predictor of death in this population, the improvement in predictive power of sudden arrhythmic death was much more substantial.
Discussion The study shows convincingly that pathological values of the 3D QRS-T angle are not only a very
Fig. 2. Areas under the curve (AUC) of receiver operator characteristics (ROC) obtained with different combinations of risk factors used to predict the cases of all cause mortality (top part) and of sudden arrhythmic death (bottom part). Open marks correspond to the risk factors listed at the horizontal axes, light full squares to the listed risk factors combined with left ventricular ejection fraction, and dark full squares to the listed risk factors combined with total cosine R-to-T. Abbreviations as in Table 1.
potent risk stratifier of post-infarction patients who are at increased risk of sudden arrhythmic death. While the predictive value of all the individual characteristics, including TCRT, is rather modest, the areas under ROCs obtained for com-
206 Journal of Electrocardiology Vol. 37 Supplement 2004 Table 3. Areas Under the Receiver Operator Characteristics Obtained for Risk Stratification All-cause mortality Fisk Factor(s)
Mean
95% Confidence Interval
LVEF TCRT LVEF ⫹ HR LVEF ⫹ HRVi LVEF ⫹ TS LVEF ⫹ TCRT LVEF ⫹ TCRT ⫹ HRVi LVEF ⫹ TCRT ⫹ TS LVEF ⫹ TCRT ⫹ HR HRVi ⫹ TCRT ⫹ TS HRVi ⫹ TCRT ⫹ HR LVEF ⫹ HRVi ⫹ TCRT ⫹ TS LVEF ⫹ HRVi ⫹ TCRT ⫹ HR
0.6610 0.6857 0.7396 0.7197 0.6984 0.7631 0.8067 0.8057 0.7988 0.7933 0.7995 0.8289 0.8359
(0.6588–0.6633) (0.6834–0.6880) (0.7377–0.7414) (0.7178–0.7215) (0.6965–0.7004) (0.7611–0.7651) (0.8049–0.8084) (0.8040–0.8073) (0.7971–0.8005) (0.7916–0.7951) (0.7978–0.8012) (0.8273–0.8304) (0.8344–0.8375)
Arrhythmic Death Mean
95% Confidence Interval
0.6346 0.7275 0.7673 0.7540 0.6902 0.8057 0.8623 0.8344 0.8456 0.8507 0.8707 0.8789 0.8938
(0.6309–0.6383) (0.7241–0.7309) (0.7645–0.7700) (0.7514–0.7566) (0.6869–0.6935) (0.8028–0.8086) (0.8602–0.8644) (0.8318–0.8370) (0.8434–0.8479) (0.8485–0.8528) (0.8687–0.8726) (0.8769–0.8809) (0.8920–0.8956)
The table lists areas under the receiver operator characteristics calculated for different combinations of risk factors used to predict the cases of death during follow-up. Abbreviations as in Table 1 All the risk factor combinations with non-overlapping 95% confidence interval of means of the areas were also statistically different in non-parametric comparisons.
binations of risk stratifiers suggest that a combination of different characteristics may improve the practicality of the identification of patients who might benefit from prophylactic intervention very considerably. The observation of the strong association of pathological values of TCRT with poor outcome is in a good agreement with previously published results (5–7). In this sense, the observations made in this study are not surprising. It is also not surprising that pathological values of TCRT are particularly associated with the risk of sudden arrhythmic death. The pathologies of ventricular re-
polarisation clearly play an important role in the genesis and pathophysiology of proarrhythmic substrates. On the contrary, the predictive value of HRV and of heart rate turbulence slope observed in this investigation was low. This is at odds with previously published observations (14 –16) and it is indeed contradicting even the data found in other sections of this Post-Infarction Survey (17,18). The particular selection of patients might have introduced some bias. Since the digital ECGs were restricted to the orthogonal XYZ leads, we were unable to evaluate
Fig. 3. Cumulative distribution of the areas under the receiver operator characteristics obtained during the repeated bootstrap experiments. Abbreviations as in Table 1. See the text for the details of the bootstrap technology.
TCRT and Post-Infarction Risk •
other morphological characteristics that are obtainable from the standard 12-lead ECG. In particular, we were unable to study the so-called T wave and QRS residua (19) that had previously been observed to be even stronger indicators of risk in ischemic heart disease (7). Although we used the bootstrap method to investigate the areas under the ROC curves statistically, all the bootstrap repetitions were taken from a moderately sized population of MI survivors with a rather low 5-year incidence of sudden arrhythmic death. At the same time, however, the difference between the LVEF-based and TCRT-based stratification of patients at risk of arrhythmic death is so striking that it might have been only little influenced by the composite of this particular population. Since we intended to use a population with a follow-up duration similar to the longevity of an ICD, the study population did not fully reflect the present trends in the therapy of acute MI. Although a high proportion of the patients were thrombolysed, the treatment with beta-blockers was lower compared to the present practice. It has been shown that the predictive power of most previously established risk stratifiers is much lower if not directly lost in patients actively treated with beta-blockers (20). Nevertheless, our most recent (unpublished, March 2004) experience with repolarization morphology suggests that TCRT is little influenced by beta-blocking therapy. In spite of these limitations, the study demonstrates the predictive value of the increased QRS-T angle convincingly. Because the predictive value of the QRS-T angle has now been consistently shown in several independent studies, and since this study shows that the pathological values of the angle are strongly associated with arrhythmic death, a sound recommendation can be made that in forthcoming trials of prophylactic ICD implantation, the 3D QRS-T angle is incorporated into the selection of patients to be treated, probably in combination with lower LVEF and possibly also with other risk characteristics.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
References 15. 1. Gregoratos G, Abrams J, Epstein AE, et al: ACC/ AHA/NASPE 2002 guideline update for implantation of cardiac pacemakers and antiarrhythmia devices: A report of the American College of Cardiology/American Heart Association Task Force on Practice Guidelines (ACC/AHA/NASPE Committee
16.
Malik, Hnatkova, and Batchvarov 207
on Pacemaker Implantation). www.acc.org/clinical/guidelines/pacemaker/pacemaker.pdf. 2002. Greenberg H, Case RB, Moss AJ, et al: MADIT-II Investigators: Analysis of mortality events in the Multicenter Automatic Defibrillator Implantation Trial (MADIT-II). J Am Coll Cardiol 43:1459, 2004 Nisam S, Henry S, Wilber DJ: Implementation of MADIT and MUSTT in clinical practice: Results of an international survey. Ann Noninvasive Electrocardiol 7:399, 2002 Myerburg RJ, Mitrani R, Interian A Jr, et al: Interpretation of outcomes of antiarrhythmic clinical trials: Design features and population impact. Circulation 97:1514, 1998 Zabel M, Acar B, Klingenheben T: Analysis of 12-Lead T-wave morphology for risk stratification after myocardial infarction. Circulation 102:1252, 2002 Kardys I, Kors JA, van der Meer IM, et al: Spatial QRS-T angle predicts cardiac death in a general population. Eur Heart J 24:1357, 2003 Zabel M, Malik M, Hnatkova K, et al: Analysis of T-wave morphology from the 12-lead electrocardiogram for prediction of long-term prognosis in male US veterans. Circulation 105:1066, 2002 Redwood SR, Odemuyiwa O, Hnatkova K, et al: Selection of dichotomy limits for multifactorial prediction of arrhythmic events and mortality in survivors of acute myocardial infarction. Eur Heart J 18:1278, 1997 Acar B, Yi G, Hnatkova K, et al: Spatial, temporal and wavefront direction characteristics of 12-lead T wave morphology. Med Biol Eng Comput 37:574, 1999 Torp-Pedersen C, Kober L, Elming H, et al: Classification of sudden and arrhythmic death. Pacing Clin Electrophysiol 20:2545, 1997 Hanley JA, McNeil BJ: The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology 143:29, 1982 Hnatkova K, Poloniecki J, Camm AJ, et al: Computation of multifactorial receiver operator and predictive accuracy characteristics. Comp Meth Prog Biomed 42:147, 1994 Efron B, Tibshirani RJ: An introduction to the bootstrap, in Monographs on Statistics and Applied Probability, vol 57, London, Chapman & Hall, 1993 Heart rate variability: standards of measurement, physiological interpretation and clinical use. Task Force of the European Society of Cardiology and the North American Society of Pacing and Electrophysiology. Circulation 93:1043, 1996. Schmidt G, Malik M, Barthel P, et al: Heart rate turbulence after ventricular premature beats as a predictor of mortality after acute myocardial infarction. Lancet 353:1390, 1999 Barthel P, Schneider R, Bauer A, et al: Risk stratification after acute myocardial infarction by heart rate turbulence. Circulation 108:1221, 2003
208 Journal of Electrocardiology Vol. 37 Supplement 2004 17. Odemuyiwa O, Malik M, Farrell T, et al: A comparison of the predictive characteristics of heart rate variability index and left ventricular ejection fraction for all-cause mortality, arrhythmic events and sudden death after acute myocardial infarction. Am J Cardiol 68:434, 1991 18. Hartikainen JEK, Malik M, Staunton A, et al: Distinction between arrhythmic and nonarrhythmic death after acute myocardial infarction based on heart rate variability, signal-averaged electrocardiogram, ventric-
ular arrhythmias and left ventricular ejection fraction. J Am Coll Cardiol 28:296, 1996 19. Malik M, Acar B, Yi G, et al: QT dispersion does not represent electrocardiographic interlead heterogeneity of ventricular repolarization. J Cardiovasc Electrophysiol 11:835, 2001 20. Huikuri H, Tapanainen JM, Lindgren K, et al: Prediction of Sudden Cardiac Death After Myocardial Infarction in the Beta-Blocking Era. J Am Coll Cardiol 42:652, 2003