Evaluation of the Spatial Aspects of T-Wave Complexity in the Long-QT Syndrome
Background The duration of the QT interval is only a gross estimate of repolarization. Besides its limited accuracy and reproducibility, it does not provide information on the morphology of the T wave; thus, morphologic alterations such as notches can be only qualitatively described but not objectively quantified.
Methods and Results To measure the complexity of repolarization in the long-QT syndrome (LQTS) patients, we previously applied principal component analysis to body surface mapping and found it useful in distinguishing normal from abnormal repolarization patterns (sensitivity, 87%). In the present study, we applied principal component analysis to 12-lead Holter recordings. The index of complexity of repolarization that we have developed (CR24h) reflects the average 24-hour complexity of repolarization and is mathematically defined as the average ratio between the second and the first eigenvalue. We studied 36 LQTS patients and 40 control subjects. A mean of 22±1.3 ECG recordings at 1-hour intervals was used in each patient, and a total of 1655 recordings were analyzed. CR24h was significantly higher in LQTS than in control subjects (34±12% versus 13±3%; P<.0001). A CR24h exceeding 2 SD above the mean of the control group (>20%) was present in 32 of 36 patients (88%). The negative predictive value of CR24h in LQTS was 88%, and the combination of prolonged QT and abnormal CR24h identified all LQTS patients from normal subjects, including 4 affected symptomatic individuals with a normal QT interval duration, suggesting that CR24h provides information independent of QT duration.
Conclusions Our data suggest that principal component analysis applied to 24-hour, 12-lead Holter recording adequately quantifies the complexity of ventricular repolarization and may become a useful noninvasive diagnostic tool in LQTS.
The presence of heterogeneity in the recovery of ventricular refractory periods is an important factor in arrhythmia development. Nonetheless, a substantial gap exists between the large amount of experimental data supporting this concept1 2 3 4 and the paucity of clinical tools to identify the presence of this substrate in patients. Evaluations of repolarization by monophasic action potential recording5 and body surface mapping6 7 8 9 have been proposed. However, they have failed to become standard in clinical practice because they require invasive studies, expensive instruments, or complex analysis. More recently, the need for a reliable and practical method to quantify ventricular repolarization seemed to have been answered by the evaluation of interlead variability of QT interval duration on 12-lead ECG recordings.10 This approach has rapidly gained popularity, and data have appeared showing that a high QT dispersion is present in individuals with a variety of heart diseases ranging from the long-QT syndrome (LQTS) to hypertrophic cardiomyopathy, cardiac failure, myocardial infarction, and angina.11 12 13 14 15 However, several investigators raised concerns about methodological aspects of the measurement of QT dispersion. It has been suggested16 17 18 that factors such as the difficulty in consistently defining the end of the T wave, the inability to measure QT interval in all leads, the presence of U waves, and notched T waves may contribute to poor interindividual and intraindividual reproducibility of QT dispersion18 and thus may reduce its power to predict the risk of arrhythmic events.19 20 21 22
Despite its well-known methodological limitations,23 the measurement of the QT interval remains a reliable, albeit gross, index for the quantification of ventricular repolarization. However it has been recognized that “duration“ is only one aspect of repolarization. Factors such as T-wave shape, the presence of notches,24 or other morphological aberrations carry major implications for the identification of “abnormal” repolarization. One illustration of the need for other measures of repolarization is the evidence that >10% of gene carriers of LQTS have a normal QT.25 All these considerations indicate the persistence of a need for a more sensitive ECG parameter that would allow a global evaluation of repolarization and provide information to augment the clinically established QT interval measurement.
The objective of the present study was to evaluate a novel ECG-based approach to study ventricular repolarization. The motivation for the approach is rooted in the appreciation, almost from the inception of ECG recording, that the normal T wave is monophasic and without notches. The method that we have used is based on principal component analysis8 9 26 applied for the first time to 12-lead as opposed to the many leads used during body surface mapping. Continuous 24-hour Holter recording allows measurement of the variability of repolarization features. Finally, compared with other methods available to characterize the QT interval, this method does not require identification of the end of the T wave. Conceptually similar to measuring the complexity of a musical chord by Fourier analysis, principal component analysis offers a quantitative means of measuring the extent to which multiple forms are required to describe the T-wave morphology. Thus, principal component analysis can be used to directly measure the “complexity” of the T wave. Using this analytical method, we studied the variability of repolarization patterns in normal subjects and in LQTS patients to define the discriminative capability of the technique among the two groups. Preliminary data have been presented elsewhere.27 28
The study population consisted of 40 healthy subjects with no heart disease, normal ECGs, and negative physical examinations and of 36 LQTS patients with either a prolonged QT interval and/or a history of syncopal episodes and a family history for premature sudden death. All the patients had a diagnostic score >4 (mean value, 5±0.9,) which corresponds to a “definite LQTS diagnosis.”29 Table 1⇓ gives the clinical characteristics of patients.
A 12-lead Holter recording (H12 recorder, Mortara Instrument) was obtained in all patients. The mean duration of the recording was 22±2 hours. Torso positions were used for the limb electrodes. Three leads (II, V1, and V5) were continuously acquired over the registration period, and 4-second segments of simultaneous 12-lead ECGs were obtained at 30-second intervals. The use of leads I and II (as opposed to, for example, unipolar leads LA and RA referenced to Wilson terminal, similar to the chest leads) is arbitrary. To test that this choice did not affect the results, the analysis was repeated using RA and LA unipolar limb leads (data shown in the “Appendix”).
Principal Component Analysis
Principal component analysis, as described by Lux et al,26 was applied to the 4-second-long 12-lead ECG recordings selected at 1-hour intervals starting from the first available ECG. A mean of 22±1.3 recordings was analyzed in each of the 76 subjects for a total of 1655 ECG recordings analyzed. Each QRST complex in the selected 4-second segment was independently analyzed, and the results were averaged over the available cycles to determine representative values for each ECG. The voltage of the extended ST-T segment was algorithmically defined as beginning at QRS offset and ending at the rate-expected end of T according to Bazett’s formula. To this expected QT, we added 0.1 second to obtain the ST-T interval for analysis. This interval is defined independently from human or automated labeling of the T-wave end. This criterion ensures inclusion of the entire repolarization waveforms even when prolonged as in LQTS. As a consequence, in normal individuals the interval analyzed by principal component analysis extends beyond the end of T wave and includes part of the TP segment. The covariance of samples in the extended ST-T interval of each QRST complex was determined, and eigenvalues were obtained. The complexity values for each QRS in the 4-second interval were then averaged to obtain a single measurement representing the 1-hour interval.
The voltage of the ST-T interval in the 12 leads was sampled every 5.5 ms, and principal component analysis was performed on these amplitudes (see the “Appendix” for calculation methods). An average 12-lead T-wave magnitude <100 μV was used as a criterion to exclude low-voltage recordings from the analysis. The principal component analysis allowed the identification of a set of eight values that represent the relative magnitude of the spatial components of repolarization. The evaluation of the relative contribution of these components provided an estimate of the spatial complexity of repolarization (CR). CR was calculated using three indexes: (1) the ratio of the second eigenvalue to the root mean square of all eight eigenvalues multiplied by 100, (2) the root mean square of the second through eighth eigenvalues divided by the root mean square of the first through eighth eigenvalues multiplied by 100, and (3) the ratio of the second eigenvalue to the first multiplied by 100. (In all cases, we have used the term “eigenvalue” to refer to the square root of the diagonal members of the diagonalized covariance matrix.)
CR was measured every hour throughout the Holter recording, and its average daily value was defined as 24-hour CR (CR24h).
QT Interval and QT Dispersion Measurement
QT interval, QTc (calculated using Bazett’s formula: QT/√ RR), QT dispersion, and QTc dispersion were manually calculated from the first 12-lead ECG recording in each patient and control subject by two independent investigators (V.P. and L.D.) who were unaware of the patient’s clinical history. The QT and RR intervals were measured by a digitizer connected to a PC using custom-made software (Experientia). The QT interval and the preceding RR interval were measured in 2 to 4 consecutive beats in at least 10 leads. The software automatically calculated QTc, QT dispersion defined as the longest (QTmax) minus the shortest (QTmin) QT interval of the 12 leads, and QTc dispersion defined as QTcmax−QTcmin.
Statistical analysis was performed by use of paired and unpaired t tests for coupled observations and Fisher’s exact test for frequency distribution. Significance was accepted at a value of P<.05. Linear regression analysis was performed to study the interdependency between variables in the different groups. Statistical analysis was performed with the SPSS package.
QTc and QTc Dispersion
The values of QT interval, QTc (measured in D2), QT dispersion, and QTc dispersion in the two groups under study are reported in Table 2⇓. As expected, QTc was significantly longer in LQTS than in control subjects (514±59 versus 414±18 ms, P<.0001). QTc dispersion was significantly higher in LQTS patients than in control subjects (82±37 versus 38±9 ms, P<.0001). The linear regression analysis showed no correlation between QTc and QTc dispersion (r2=.04). A QTc exceeding by more than 2 SD (450 ms) the mean value observed in control subjects was present in 32 of 36 LQTS patients (sensitivity, 88%). A QTc dispersion exceeding the mean+2 SD (56 ms) of the value observed in control subjects was present in 26 of 36 LQTS patients (sensitivity, 72%).
Complexity of Repolarization
The CR24h was significantly higher in LQTS patients than in control subjects using any of the three indexes (data are presented in Table 2⇑). We elected to perform further analysis using the ratio between the second and the first eigenvalue (Fig 1⇓). A linear regression analysis showed no significant correlation between CR and QTc or QT dispersion in the two groups.
A CR24h exceeding the mean+2 SD (mean+2 SD=19%) of the value observed in control subjects was present in 32 of 36 LQTS patients (sensitivity, 88%).
CR24h was higher in the 26 LQTS patients with visible notches on ECG than in the 10 patients with a smooth T-wave profile (37±11% versus 26±10%, respectively; P<.005). There was no significant difference in CR24h between the 27 patients treated with β-blockers and the 9 patients without therapy at the time of the study. Similarly, no significant difference was observed between patients with history of syncope or cardiac arrest (n=29) versus the asymptomatic individuals (n=7). No correlation was found between the LQTS diagnostic score29 (mean score in the group, 5.1±0.97; range, 4 to 9) and CR24h. Thus, CR24h correlated with the presence of notches on the T wave but not with therapy, symptoms, or diagnostic score.
Four LQTS patients presented with a normal QT interval (QTc=430, 413, 417, and 437 ms, respectively). In all of them, CR24h was above normal values (CR24h=41%, 57%, 25%, and 22%, respectively). In these individuals, QTc dispersion was above normal values only in 2 of the 4 patients (QTc dispersion=42, 163, 194, and 46 ms, respectively; Fig 2⇓).
When CR24h is measured as the coefficient of variability (CV) of CR24h (CVCR24 h, defined as the SD of the hourly recorded CR ratio divided by the mean CR over the 24 hours multiplied by 100), LQTS patients have a higher CVCR24h than control subjects (CVCR24h, 35±10 versus 24±7, respectively; P<.0001). An example of the 24-hour variability in 1 control subject and in 1 LQTS patient is shown in Fig 3⇓.
One patient presented an episode of T-wave alternans during Holter recording. Principal component analysis applied to the T waves of the alternating beats showed different complexity of the two patterns of repolarization (Fig 4⇓).
Diagnostic Value of Dynamic Complexity of Repolarization
The negative predictive value (NPV) of QTc, QTc dispersion, and CR24h and of the combination of QTc and CR24h, QTc, and QTc dispersion, and CR24h and QTc dispersion are shown in Table 3⇓. QTc and CR24h have the same NPV (91%) and perform better than QTc dispersion (NPV, 80%; P=.09); the two parameters that when combined achieve the best performance in the identification of LQTS patients are QTc and CR24h (NPV,100%). When QTc and QTc dispersion are combined, NPV increases (from 91% with QTc alone to 97% with QTc plus QTc dispersion), yet not all patients are identified (Table 3⇓).
Diagnostic Value of CR on a Single 12-Lead Recording
When CR is calculated by applying principal component analysis to a single 12-lead ECG recording (in all patients, by convention, the first episode included in the Holter evaluation was selected), the ratio between the second and the first component was significantly higher in LQTS than in control subjects (30±15% versus 12±4.5%, P<.0001). However, overlap between the two groups was present. Thus, both the sensitivity and the NPV of complexity on a single ECG measurement were lower compared with that measured during the 24 hours, ie, the average 24-hour CR (sensitivity, 69% versus 88%; Fisher’s exact test, P<.05; NPV, 78% versus 91%; Fisher’s exact test, P=.08).
In the present study, principal component analysis was used in healthy control subjects and in LQTS patients because these two groups are thought to represent the two opposite ends of the spectra of ventricular repolarization patterns, from normal to most abnormal.
The first major finding of this study is that it is possible to quantify the dynamic behavior of T-wave complexity by applying principal component analysis26 to 12-lead Holter recordings. The second major finding is that this approach differentiates LQTS patients from control subjects with 88% sensitivity. When the duration of the QT interval and complexity are combined, they correctly identify all LQTS patients from control subjects, thus allowing ECG diagnosis even in those patients (11%) with normal QT intervals. This algorithmically defined parameter provides clinical information independent of QTc measurement, and it significantly increases the ECG diagnostic capability in LQTS.
Quantification of Ventricular Repolarization
Several methods are available at the experimental level to quantify abnormalities of repolarization5 6 8 ; however, few of these techniques are suitable for routine clinical use. Thus, the assessment of ventricular repolarization is still based largely on QT and QTc measurement and on a qualitative description of morphological alterations such as the presence of notched, bifid, or biphasic T waves.24 A set of new morphological ECG parameters proposed by Merri et al30 and by Benhorin et al31 could provide a better description of repolarization and be more reproducible than QT interval duration. Unfortunately, these indexes have not yet obtained widespread application in clinical practice.
In the attempt to provide a new and more powerful method to quantify repolarization, Day et al10 proposed as the index of abnormal repolarization the interlead variability of QT defined as the difference between the longest and shortest QT interval measured on a standard 12-lead ECG. This method is based on the correct attribution of the boundaries of the QT interval. It is therefore unavoidably affected by the same limitations of the QT measurement23 32 such as the difficulty in defining the end of the T wave, the confounding role of U waves or biphasic shapes of the T wave, signal gain, and paper speed. These limitations are likely to increase the interobserver variability of QT dispersion and reduce its reproducibility.16 17 18 Additionally, both the QT interval and QT interval dispersion reflect a prolongation of the repolarization without taking into account the alterations occurring during the temporal development of the T wave (eg, notches). The LQTS patients, who are often considered a clinical model of markedly abnormal repolarization, may have a normal duration of the QT interval25 33 and yet present with bifid and notched T waves.24 Despite the qualitative nature of the notches, their importance is recognized by the inclusion of this morphological alteration in the diagnostic criteria of LQTS.29 In the “score” systems, they receive a value identical to that of a positive family history for LQTS.
From this brief overview, it follows that the methods in current clinical use to measure ventricular repolarization fail to provide a quantitative assessment of T-wave morphology.
A novel method for a clinically useful determination of ventricular repolarization should integrate information obtained from several recording sites (spatial dispersion) and should be assessed on multiple samples to take into account time-dependent changes (24-hour variability). We propose that principal component analysis applied to the 12-lead Holter recording accomplishes these objectives.
Principal Component Analysis as an Index of CR
Principal component analysis has already been applied to the study of repolarization during body surface mapping in LQTS patients. We have previously shown8 9 that this method of analysis, applied to body surface recording from 117 electrodes, distinguishes between normal and abnormal repolarization patterns with a sensitivity of 87%. Because body surface mapping is a complex technique limited to research applications, we attempted the same approach by using a standard 12-lead ECG recording. The information derived from principal component analysis separates and quantifies the relative weight of the orthogonal components of repolarization. When repolarization is uniform, as happens in normal individuals, one major spatial component (eigenvector) can be identified; conversely, when the repolarization pattern becomes fragmented, the relative value of the smaller vectors increases proportionally. Such an approach allows a comparison between the morphology of the T wave across the 12 leads and quantification of T-wave abnormalities in an observer-independent way.
On the basis of the present findings, principal component analysis applied to the 12-lead ECG accurately distinguishes normal from abnormal repolarization patterns and allows identification of LQTS patients from normal subjects with a sensitivity of 88%, a value remarkably identical to the 87% obtained 10 years ago by our group with body surface mapping.8
No correlation was found between QTc or QTc dispersion and the 24-hour CR. Therefore, the principal component analysis provides information different from that obtained by interlead dispersion or duration of QT.
When the QTc is used in the diagnosis of LQTS, it is highly specific because a markedly prolonged QT is virtually absent in the normal population. However, its sensitivity is suboptimal because gene carriers can present with a modest and even normal QT interval.25 The present data support the view that even LQTS patients with a normal QT duration have an abnormal repolarization pattern that is not detected with QT interval and QTc measurement. The present data demonstrate that the evaluation of the dynamic complexity may help in detecting these more subtle abnormalities and that it may become a useful diagnostic tool for those LQTS patients currently misdiagnosed by a traditional clinical evaluation based on QTc measurement.
Variability of Repolarization in LQTS
We previously reported that in LQTS the reproducibility over 24 hours of the QT interval and QT dispersion is low18 ; we found a mean variability of ±20 ms in QTc and ±40 ms in QTc dispersion. The 24-hour CV of repolarization is very pronounced in LQTS patients compared with control subjects. In a preliminary study of CR,28 we demonstrated that in 23 patients after myocardial infarction, the T-wave complexity is high (mean value of CR24h=25±11), yet the 24-hour variability is low (CR24h=24±8). In agreement with these data is the observation that the diagnostic power of principal component analysis is largely reduced when it is applied to a single as opposed to multiple recordings in these two groups of control subjects and LQTS patients. As shown in Fig 3⇑, there are times when complexity in the LQTS patient and the control subject overlaps; therefore, measurement of complexity in a single recording could grossly underestimate the presence of abnormal repolarization patterns. These observations suggest that the assessment of repolarization in LQTS should include at least multiple samples within a 24-hour period. Whether sensitivity of QTc and QTc dispersion in identifying abnormal repolarization in LQTS would increase if these two indexes were calculated over 24 hours remains to be assessed.
Applications of Principal Component Analysis Applied to ECG Recording
We showed that principal component analysis provides a measure of T-wave complexity that allows correct identification of patients with abnormal repolarization. The 24-hour variability observed in LQTS patients suggests that the method not only portrays the “substrate” responsible for the abnormal repolarization but also reflects changes mediated by trigger factors. This concept is further supported by the evidence that during T-wave alternans,33 a marker of electrical instability, the complexity of repolarization is increased and alternates on a beat-to-beat basis.
It is therefore tempting to speculate that principal component analysis applied to standard 12-lead ECGs may help in identifying patients with abnormal repolarization at higher risk of ventricular arrhythmias such as, for example, those patients treated with repolarization-prolonging agents at higher risk of torsade de pointes.34
The present data support the proposal of a new approach, based on the application of principal component analysis8 9 26 to 24-hour 12-lead Holter recordings, for noninvasive quantification of ventricular repolarization. This approach provides several advantages over the existing methods for the quantification of repolarization: (1) produces an index that provides information on spatial dispersion of repolarization; (2) can be assessed over time, providing information on the dynamicity of complexity of repolarization; (3) is independent of the subjective definition of the QT interval duration; and (4) is entirely automatic and does not require repetitive and time-consuming calculations.
When applied to LQTS patients, this method distinguishes between affected and nonaffected individuals. Principal component analysis provides an accurate estimate of the homogeneity of repolarization and may become diagnostic in patients affected by LQTS with a borderline QT interval.
Principal Component Analysis as Applied to 12-Lead ECG Recording
Let the indices λ and μ designate one of the eight leads I, II, V1 through V6; κ, one of the sample times in the QT interval (described above as the interval from the end of QRS to beyond the end of T); and y, a sample amplitude referenced to a baseline drawn from QRS onset to the end of the QT interval in each lead. A spatial covariance matrix, Rλμ, is calculated from yκλ as: where N is the number of samples in the QT interval. The covariance matrix R has eigenvalues and eigenvectors obtained, for example, by singular value decomposition of R.
The eigenvectors correspond to principal axes in the eight-dimensional space of leads, and the eigenvalues correspond to the mean square amplitude of the signal vector yκλ projected onto each of these axes. The first principal component and axis correspond to the square root of the largest eigenvalue and its eigenvector, the second principal component to the square root of the next largest eigenvalue, and so on, through the eighth principal component. Note that for each subject studied, an individual covariance matrix is computed from which the individual ratio of the principal components is derived.
The use of leads I and II (as opposed to, for example, unipolar leads LA and RA referenced to the Wilson terminal, similar to the chest leads) is arbitrary. Confirmation that this choice did not affect the results, either quantitatively or qualitatively, was made by repeating the analysis using unipolar (RA and LA) limb leads. Comparative results for leads I and II and for unipolar leads RA and LA are reported in Table 4⇓.
This study was partially supported by CEE grant BMH4-CT96-0028. Dr Priori was partially funded by a Bristol-Myers-Squibb research grant. We wish to thank Dr Maurizio Fumagalli for technical assistance and Pinuccia De Tomasi for editorial support.
Reprints requests to Silvia G. Priori, MD, PhD, Molecular Cardiology, Fondaz S Maugeri, Via Ferrata 8, 27100 Pavia, Italy.
- Received March 28, 1997.
- Revision received May 27, 1997.
- Accepted June 6, 1997.
- Copyright © 1997 by American Heart Association
Surawicz B. The Electrophysiologic Basis of ECG and Cardiac Arrhythmias. Baltimore, Md: Williams and Wilkins; 1995.
Abildskov JA, Burgess MJ, Lux RL, Wyatt RF. Experimental evidence for regional cardiac influence in body surface isopotential maps of dogs. Circ Res. 1976;38:386-391.
Han J, Moe GK. Nonuniform recovery of excitability in ventricular muscle. Circ Res. 1964;4:44-60.
Kuo CS, Munakata K, Reddy CP, Surawicz B. Characteristics and possible mechanisms of ventricular arrhythmia dependent on the dispersion of action potential durations. Circulation. 1983;67:1356-1367.
Bonatti V, Rolli A, Botti G. Recording of monophasic action potentials of the right ventricle in long QT syndromes complicated by severe ventricular arrhythmias. Eur Heart J. 1983;4:168-179.
De Ambroggi L, Bertoni T, Locati E, Stramba-Badiale M, Schwartz PJ. Mapping of body surface potentials with the idiopathic long QT syndrome. Circulation. 1986;74:1334-1345.
Day CP, McComb JM, Campbell RWF. QT dispersion: an indication of arrhythmic risk in patients with long QT intervals. Br Heart J. 1990;63:342-344.
Higham PD, Furniss SS, Campbell RWF. QT dispersion and components of the QT interval in ischemia and infarction. Br Heart J. 1995;73:32-36.
Buja G, Miorelli M, Turrini P, Melacini P, Nava A. Comparison of QT dispersion in hypertrophic cardiomyopathy between patients with and without ventricular arrhythmias and sudden death. Am J Cardiol. 1993;27:973-976.
Dritsas A, Puri S, Davis G, Krikler S, Cleland J, Nihoyannopoulos CM, Oakley CM. QT dispersion is increased in hypertrophic cardiomyopathy compared with secondary left ventricular hypertrophy. Eur Heart J. 1993;14(suppl):212. Abstract.
Priori SG, Napolitano C, Diehl L, Schwartz PJ. Dispersion of the QT interval: a marker of therapeutic efficacy in the idiopathic long QT syndrome. Circulation. 1994;89:1681-1689.
Hnatkova K, Malik M, Kautzner J, Gang Y, Camm AJ. Adjustment of QT dispersion assessed from 12-lead electrocardiograms for different numbers of analyzed electrocardiographic leads: comparison of stability of different methods. Br Heart J. 1994;72:390-396.
Priori SG, Diehl L, Napolitano C, Paganini V, Schwartz PJ. Criteria to assess intervention-related changes in QT and QT dispersion: a study in subjects with normal and prolonged QT intervals. Circulation. 1995;92(suppl I):I-681. Abstract.
Barbato G, Franco N, di Niro M, Pavesi PC, Bracchetti D. Correlation between dispersion of ventricular repolarization and malignant arrhythmias during acute myocardial infarction. Eur Heart J. 1995;16(suppl):446. Abstract.
Kautzner J, Kishore R, Camm AJ, Malik M. The role of QT-dispersion for prediction of ventricular arrhythmias after myocardial infarction. Eur Heart J. 1995;16(suppl):135. Abstract.
Murray A, McLaughlin NB, Bourke JP, Doig JC, Furniss SS, Campbell RWF. Errors in manual measurement of QT intervals. Br Heart J. 1994;71:386-390.
Lux RL, Evans AK, Burgess MJ, Wyatt RF, Abildskov JA. Redundancy reduction for improved display and analysis of body surface potential maps, I: spatial compression. Circ Res. 1981;49:186-191.
Priori SG, Napolitano C, Diehl L, Paganini V, Fumagalli M, Mortara D, Schwartz PJ. Principal component analysis of repolarization: a novel index of complexity of ventricular repolarization in the long QT syndrome. J Am Coll Cardiol. 1996;27:21A. Abstract.
Napolitano C, Diehl L, Paganini V, Cantù F, Fumagalli M, Mortara D, Priori SG. Principal component analysis identifies abnormal complexity of repolarization in post myocardial infarction patients. J Am Coll Cardiol. 1996;27:121A. Abstract.
Schwartz PJ, Moss AJ, Vincent GM, Crampton RS. Diagnostic criteria for the long QT syndrome: an update. Circulation. 1993;88:782-784.
Merri M, Benhorin J, Alberti M, Locati E, Moss AJ. Electrocardiographic quantitation of ventricular repolarization. Circulation. 1989;80:1301-1308.
Benhorin J, Merri M, Alberti M, Locati E, Moss AJ, Hall WJ, Cui L. Long QT syndrome. New electrocardiographic characteristics. Circulation. 1990;82:521-527.
Lepeschkin E, Surawicz B. The measurement of the QT interval of the electrocardiogram. Circulation. 1953;4:378-388.
Thomas SHL. Drugs and the QT interval. Adverse Drug Reaction Bull. 1997;182:691-694.