Author + information
- Received July 5, 2016
- Revision received October 18, 2016
- Accepted October 20, 2016
- Published online April 17, 2017.
- Andreu Porta-Sánchez, MDa,
- David R. Spillane, MSc, MDCM(c)b,
- Louise Harris, MDa,
- Joel Xue, PhDc,
- Pat Dorsey, BSc,
- Melanie Care, MScc,
- Vijay Chauhan, MDa,
- Michael H. Gollob, MDa and
- Danna A. Spears, MDa,∗ ()
- aDivision of Cardiology, Peter Munk Cardiac Center, University Health Network, Toronto, Ontario, Canada
- bFaculty of Medicine, McGill University, Montreal, Quebec, Canada
- cGE Healthcare, Wauwatosa, Wisconsin
- ↵∗Address for correspondence:
Dr. Danna A. Spears, Department of Cardiology, Peter Munk Cardiac Center, 200 Elizabeth Street, Toronto, Ontario M5G 2C4, Canada.
Objectives This study aims to assess the capability of T-wave analysis to: 1) identify genotype-positive long QT syndrome (LQTS) patients; 2) identify LQTS patients with borderline or normal QTc interval (≤460 ms); and 3) classify LQTS subtype.
Background LQTS often presents with a nondiagnostic electrocardiogram (ECG). T-wave abnormalities may be the only marker of this potentially lethal arrhythmia syndrome.
Methods ECGs taken at rest in 108 patients (43 with LQTS1, 20 with LQTS2, and 45 control subjects) were evaluated for T-wave flatness, asymmetry, and notching, which produces a morphology combination score (MCS) of the 3 features (MCS = 1.6 × flatness + asymmetry + notch) using QT Guard Plus Software (GE Healthcare, Milwaukee, Wisconsin). To assess for heterogeneity of repolarization, the principal component analysis ratio 2 (PCA-2) was calculated.
Results Mean QTc intervals were 486 ± 50 ms (LQTS1), 479 ± 36 ms (LQTS2), and 418 ± 24 ms (control subjects) (p < 0.05). MCS and PCA-2 differed between LQTS patients and control subjects (MCS: 117.8 ± 57.4 vs. 71.9 ± 16.2; p < 0.001; PCA-2: 20.2 ± 10.4% vs. 14.6 ± 5.5%; p < 0.001), LQTS1 and LQTS2 patients (MCS: 96.3 ± 28.7 vs. 164 ± 75.2; p < 0.001; PCA-2: 17.8 ± 8.3% vs. 25 ± 12.6%; p < 0.001), and between LQTS patients with borderline or normal QTc intervals (n = 17) and control subjects (MCS: 105.7 ± 49.9 vs. 71.9 ± 16.2; p < 0.001; PCA-2: 18.1 ± 7.2% vs. 14.6 ± 5.5%; p < 0.001). T-wave metrics were consistent across multiple ECGs from individual patients based on the average intraclass correlation coefficient (MCS: 0.96; PCA-2: 0.86).
Conclusions Automated T-wave morphology analysis accurately discriminates patients with pathogenic LQTS mutations from control subjects and between the 2 most common LQTS subtypes. Mutation carriers without baseline QTc prolongation were also identified. This may be a useful tool for screening families of LQTS patients, particularly when the QTc interval is subthreshold and genetic testing is unavailable.
Congenital long QT syndrome (LQTS) is an inherited disorder of myocardial repolarization, conferring an increased risk of sudden cardiac death. The most common subtypes are caused by mutations in genes KCNQ1 (LQT1), KCNH2 (LQT2), and SCN5A (LQT3) (1) that encode cardiac potassium (Kv7.1, Kv11.1) and sodium ion channels (NaV1.5). The diagnostic probability of LQTS is high when a QTc interval at rest exceeds 480 ms. However, QTc measurements are variable in LQTS, and it is estimated that 30% to 50% of patients with LQTS1, 2, and 3 present with a nondiagnostic QTc at rest (2). The identification of other distinguishing electrocardiographic (ECG) features aid the diagnosis of LQTS in such cases.
LQTS is associated with abnormal T-wave morphology (3–5), and qualitatively described T waves have been identified for each major LQTS subtype (6). LQTS1 patients (decreased slow component of the delayed rectifier potassium current) tend to have early-onset, broad-based T waves. LQTS2 patients (decreased human ether-a-go-go-related gene, delayed rectifier potassium current) commonly have asymmetric, low amplitude, bifid, or notched T waves. LQTS3 patients (increase in late sodium current INa) have long isoelectric ST- segments with late-appearing normal morphology T waves, but these T-wave features are often subtle and can be easily overlooked by a nonexpert in the field of inherited arrhythmias. Thus, an automated algorithm that could potentially help in the diagnosis of LQTS patients may be of great importance.
Diagnosis of LQTS using ECG analysis may be particularly valuable when options for genetic testing are not available or in borderline phenotypes. Accurately differentiating the subtype of LQTS could have implications in the choice of beta-blockers and the use of other medications (7–11), and in family cascade screening when genetic testing is not readily available.
In the era of digital ECG, T-wave morphology can now be evaluated quantitatively using morphological features, including flatness, asymmetry, and notching (12). In the present study, we investigated whether a novel software-based, T-wave analysis could accurately discriminate gene-positive LQTS patients from control subjects. We also examined whether the software was able to identify LQTS patients whose ECGs did not show QTc prolongation and whether the features of the T-wave analysis allowed for classification of LQTS subtypes.
Patients were retrospectively studied in a single institution (Toronto General Hospital, University Health Network, Toronto, Ontario, Canada). These patients were identified by review of the clinical genetics database at this institution. Only patients who were not on QT prolonging agents and who had no reversible causes of QTc prolongation at clinical presentation were included in the analysis. Most of the included patients were untreated for LQTS when the first ECG was obtained. The protocol was approved by the Research Ethics Board at the Toronto General Hospital. LQTS patients had at least 1 digitally captured ECG available and were gene-positive. The control group included 45 healthy subjects with normal ECGs who were studied in a general cardiology clinic and evaluated for symptoms that were deemed noncardiac in origin, and who were found to have a normal ECG with a normal 2-dimensional echocardiogram.
In all LQTS patients, the first available digital 12-lead ECG recording at rest was used for analysis. The T-wave morphology parameters were generated by QT Guard Plus Software (GE Healthcare, Milwaukee, Wisconsin). Although the details of the computation of T-wave flatness, asymmetry, and the presence of notching have been described elsewhere (13), a brief explanation follows. A median beat was derived from each independent lead (leads I to II and V1 to V6) of the 10-s, 12-lead ECG, and the morphology measures were assessed in the first eigenvector of principal component analysis (PCA) of 8 independent median beats. Therefore, increasing PCA values reflected the increasing flatness of the T-wave (dimensionless units). T-wave flatness is based on the inverse kurtosis (1-kurtosis) of the unit area of the T-wave. The asymmetry score (dimensionless units) evaluates differences in the slope profile and duration of the ascending and descending parts of the T-wave, whereas low values correspond with more symmetric T waves and higher values correspond with more asymmetric T waves. The presence of notching was obtained from the inverse, signed radius of curvature of the T wave, where values >0 correspond with the presence of notching. The morphology combination score (MCS) was calculated from these measures using the following equation: MCS = 1.6 × flatness + asymmetry + notch. The coefficients were calculated using the linear regression method. The T-wave offset was calculated based on a composite signal formed from all 12 lead median representative beats (14). Then the first difference of that composite signal was calculated. The T-wave offset was determined in the last segment of a T-wave in which a window of 6 ms of signal had a <2% contribution to the total T-wave area. The QT was corrected with Bazett’s formula (QT/√RR). The PCA-ratio 2 (PCA-2) was defined as the ratio of principal component 2 to principal component 1. Individuals with at least 5 ECG tracings over time were analyzed to determine the reproducibility of all the parameters measured.
Normally distributed continuous variables were compared using Student t test or analysis of variance. Continuous variables that were not normally distributed were compared using a Mann-Whitney U test for 2 group comparisons, and the Kruskal-Wallis test was used for multiple group comparisons. The p values for multiple comparisons were corrected using Bonferroni’s correction. A chi-square test was used to compare proportions. A nonparametric Fisher test was used to compare proportions that were not suitable for chi-square testing. Reproducibility of repeat measurements was assessed based on the intraclass correlation coefficient for any subject with ≥5 repeated ECGs over follow-up. A logistic regression model was used to compare the receiver-operating characteristic (ROC) curves.
Patient characteristics are summarized in Table 1. Our cohort of genetically confirmed LQTS patients consisted of 63 patients: 43 with LQTS1 and 20 with LQTS2. The ages of these patients ranged from 16 to 83 years, with 8 patients who were 18 years of age or younger. A group of 45 healthy control subjects was used for comparison. The age range of control subjects was 13 to 56 years, and 5 were 18 years of age or younger. Baseline characteristics were similar for patients and control subjects, and are summarized in Table 1. Conventional ECG parameters were similar in both LQTS patients and control subjects, except for the QTc interval (484 ± 45 ms vs. 419 ± 24 ms, respectively; p < 0.05). The mean QTc interval was not significantly different between LQTS1 and LQTS2 patients (486 ± 49 ms vs. 479 ± 36 ms; p = NS). Eleven LQTS patients had ≥5 ECGs acquired over time, which were used to calculate intraclass correlation. These patients were on beta-blocker therapy at follow-up.
T-wave morphology parameters in LQTS patients
Although the T-wave axis and QT dispersion were not significantly different in patients and control subjects (data not shown), morphology parameters showed a statistically significant difference between these groups, and between LQTS1 and LQTS2 patients (Table 2). The T-wave morphology parameters did not show collinearity with the PCA-2, and thus characterized unique features of the T wave that could be missed with the PCA.
Mean flatness, asymmetry, and notching scores were increased in LQTS patients compared with control subjects (flatness: 50.0 ± 10.2 vs. 37.7 ± 8.2; p < 0.001; asymmetry: 24.5 ± 22.5 vs. 10.8 ± 7.9; p < 0.001; notching: 2.2 ± 10.9 vs. 0 ± 0; p < 0.01) and in LQTS2 patients compared with LQTS1 patients (flatness: 55.7 ± 10.1 vs. 47.2 ± 9.1; p < 0.01; asymmetry: 39.2 ± 32.3 vs. 17.7 ± 11.1, p < 0.05; notching: 5.5 ± 18.2 vs. 0.67 ± 4.4; p < 0.001). Similarly, the MCS, calculated from these 3 variables, was significantly increased in LQTS patients versus control subjects (117.8 ± 57.4 vs. 71.9 ± 16.2; p < 0.001), particularly in LQTS2 patients compared with LQTS1 patients (164 ± 75.2 vs. 96.3 ± 28.7; p < 0.001). Figure 1 shows 4 examples of patients with different QTc values, T-wave scores, and PCA-2 values.
Mean PCA-2 scores were higher in LQTS patients compared with control subjects (20.2 ± 10.4% vs. 14.6 ± 5.5%; p < 0.001), and in LQTS2 patients compared with LQTS1 patients (25.0 ± 12.6% vs. 17.8 ± 8.3%; p < 0.01).
T-wave morphology parameters in patients with normal QTc intervals
Seventeen LQTS patients had a normal QTc interval (≤460 ms) (12 LQTS1 and 5 LQTS2 patients). A detailed analysis of these patients is presented in Table 3.
Mean flatness and asymmetry scores were significantly higher in LQTS patients with normal QTc intervals compared with control subjects (flatness: 46.5 ± 7.07 vs. 37.7 ± 8.15; p < 0.001; asymmetry: 18.9 ± 13.1 vs. 10.8 ± 7.9; p = 0.007). MCS was also significantly greater in LQTS patients with a QTc interval of ≤460 ms compared with control subjects (105.7 ± 49.9 vs. 71.9 ± 16.2; p < 0.001). Flatness, asymmetry, notching, and MCS were not significantly different between LQTS1 and LQTS 2 patients with normal QTc intervals.
Mean PCA-2 was increased in LQTS patients with normal QTc intervals compared with control subjects (18.1 ± 7.2% vs. 14.6 ± 5.5%; p < 0.001) and in LQTS2 patients compared with LQTS1 patients (24.4 ± 8.8% vs. 14.9 ± 3.3%; p = 0.007).
Diagnostic performance of MCS and PCA-2
An MCS score of ≥85 was associated with a sensitivity of 79.0%, a specificity of 82.6%, and a global accuracy of 80.6% (ROC area 0.878) (Figure 2). In LQTS patients with a normal QTc, an MCS threshold of 85 maintained a sensitivity of 72.2%, with a specificity of 82.6%, and a global diagnostic accuracy of 79.7% (ROC area of 0.817). The distribution of MCS values in these patients is shown in Figure 3.
Using a minimum PCA-2 of 15.8% as a diagnostic threshold for LQTS, a sensitivity of 59.7%, a specificity of 60.9%, and a global diagnostic accuracy of 60.2% were achieved (ROC area 0.688). When applied to LQTS patients with a normal QTc, this cutoff was associated with a sensitivity of 61.1%, specificity of 60.9%, and a poor global diagnostic accuracy of 60.9% (ROC area 0.657).
A logistic regression model was used to compare the ROC curves of both measurements. The MCS score performed significantly better (ROC area of 0.88 vs. ROC area of 0.69; p = 0.002) than the PCA-2 score.
Reproducibility of T-wave morphology analysis
Reproducibility of the parameter measurement was assessed by comparing T-wave morphology measurements produced from different ECGs of individual LQTS patients using intraclass correlation. The average intraclass correlation coefficient for each parameter was as follows: MCS: 0.96 (95% confidence interval [CI]: 0.91 to 0.99); flatness score: 0.95 (95% CI: 0.88 to 0.98); asymmetry score: 0.89 (95% CI: 0.74 to 0.97); notch score: 0.85 (95% CI: 0.64 to 0.95); and PCA-2: 0.86 (95% CI: 0.68 to 0.96).
In the present study, we analyzed the diagnostic potential of automated, quantitative analysis of T-wave morphology applied to ECGs from healthy control subjects and genotype-confirmed LQTS patients. The main findings were as follows: 1) software-based measures of T-wave morphology, including T-wave flatness, asymmetry, notching, MCS, and PCA-2, differed significantly between control subjects and LQTS patients; 2) T-wave morphology analysis further differentiated between LQTS1 and LQTS2 patients; and 3) these measured parameters allowed for accurate differentiation between LQTS patients with a QTc interval of ≤460 ms and healthy control subjects.
Automated quantification of T-wave morphology correctly differentiates LQTS from controls subjects
The presence of an abnormal T-wave morphology is recognized as one of the important ECG features of LQTS, and the earliest diagnostic criteria included the presence of notched, biphasic T waves on the ECG in a careful description by Schwartz et al. (15). Attempts to quantify abnormal repolarization have been numerous. Benhorin et al. (4) measured T-wave properties including early onset, duration, rate of repolarization, and amplitude in digital ECGs from patients with LQTS. T-wave notching was later quantified by Malfatto et al. (5), who showed the possible prognostic value of these abnormalities in LQTS patients. Later investigations measured T-wave changes that were more easily observed by the clinician—duration and amplitude (or flatness) (6), and symmetry (16). These factors were of special importance in diagnosing patients without overt QTc prolongation; however, this method carried an important risk of misclassification.
In light of the need for a standardized approach to measure T-wave morphology, automated mathematical strategies to assess myocardial repolarization abnormalities were designed. Some methods focused on the assessment of repolarization complexity, such as PCA. In brief, PCA is a mathematical method to quantitatively assess the extent to which multiple shapes are required to describe the T-wave morphology. Using this methodology in 24-h Holter recordings and creating a complexity of repolarization index, Priori et al. (17) were able to correctly differentiate LQTS patients from control subjects. However, this index did not gain widespread use, and those results have not been further validated.
Signal processing models to analyze the morphology of the T wave were then implemented, including models that quantified repolarization, such as the Hill equation proposed by Kanters et al. (18). Although those parameters were seen to differentiate LQTS1 and LQTS2 patients reliably, this method did not receive widespread clinical acceptance. Andersen et al. (13) studied 3 different T-wave morphology parameters (flatness, asymmetry, notch) in a cohort of LQTS2 patients, and found them to be significantly different from control values, and later determined that those parameters could also identify drug-induced LQTS with good accuracy (19). These T-wave changes were elegantly correlated with the degree of hERG blockade in a study by Vicente et al. (20). These previous approaches were marred by a complex analysis in the absence of automated, digitalized processing, which precluded clinical application in routine practice.
In the present study, automated analysis of T-wave morphology distinguished patients with congenital LQTS from healthy control subjects. The results of T-wave analysis were consistent across multiple ECGs from single subjects. The strengths of automated T-wave scoring reside in its easy applicability and diagnostic accuracy as complements to automated measurements currently performed in standard ECGs. These findings support the use of this technology in clinical environments where genetic testing is not available. Further studies should focus on validating this method in pediatric subjects because few subjects were younger than 18 years of age.
T-wave analysis differentiates LQTS subtypes
Diagnostic subtyping of LQTS is important for counseling of lifestyle modifications (exercise, alarm clock use) and beta-blocker choices (7–11). Conclusively determining the subtype relies on genotyping, which may present challenges with respect to access, cost, allelic heterogeneity, and variable penetrance. It has been shown in some studies that there are some patterns of T-wave morphology that are more commonly seen in each LQTS subtype (6). Our study demonstrated that automated T-wave analysis was able to differentiate between LQTS1 and LQTS 2, the most common forms of LQTS. The usefulness of this approach was also applicable in the setting of a borderline QTc, in which PCA-2 can distinguish LQTS 1 and LQTS2. In addition, this might be useful for identifying LQTS2-like T-wave patterns that may correlate with increased risk of torsade de pointes in patients with exposure to certain drugs (21).
T-wave analysis identifies LQTS patients with normal QTc intervals
It is well recognized that gene-positive LQTS patients can present with normal or near-normal QTc intervals (22). Because of the importance of identifying these individuals, this creates a diagnostic challenge, particularly in areas where genetic testing is not available. Bazett’s formula complicates matters because it is known to overestimate QTc intervals at heart rates >85 beats/min (23), a finding that is not uncommon at clinical presentation. Miscalculation and/or misinterpretation of QTc values is a major contributor to incorrectly diagnosing LQTS (24). The advantage of the software applied in this study was that the digitally measured T-wave morphology scores were independent of heart rate (25) and did not rely on identifying the end of the T-wave, making their measurement clinically useful and potentially less prone to variability.
In this study, we found that T-wave morphology, based on MCS and PCA-2, differed significantly between LQTS patients with borderline QTc intervals and control subjects. These measures also proved to be independent of QT duration, because our data indicated the ability to distinguish gene-positive LQTS patients with QTc intervals of ≤460 ms from age- and sex-matched healthy control subjects.
This study did not evaluate the relationship between T-wave morphology and outcome as it was underpowered, but his would be an important direction of future study. In addition, these results can only be extrapolated to an adult population as we did not include pediatric patients, where application of this technology may have its greatest utility.
In this study, T-wave morphology analysis using automated measurement accurately discriminated gene-positive LQTS patients from control subjects with good sensitivity and specificity. Analysis of automated T-wave morphology in this patient cohort could discriminate between the most common LQTS subtypes in patients with LQTS. In a subset of patients with QTc interval of ≤460 ms, this analysis remained useful in identifying gene carriers, and 1 parameter, PCA-2, was useful in discriminating genotype. There was a high degree of reproducibility between repeated measurements of all T-wave morphology parameters in subjects with multiple ECGs. Automated T-wave morphology analysis has the potential to be a new diagnostic tool for identifying and evaluating LQTS and may be of significant value in clinical settings where genetic testing is not feasible.
COMPETENCY IN MEDICAL KNOWLEDGE: The primary diagnostic criterion in LQTS is a prolonged QTc interval. However, T-wave abnormalities have also been found. Automated T-wave morphology analysis reproducibly quantifies the degree of T-wave morphology abnormalities with good diagnostic efficacy. This may be a useful new tool for assisting in LQTS diagnosis, particularly when the QTc interval is nondiagnostic and when genetic testing is unavailable.
TRANSLATIONAL OUTLOOK: Further studies comparing the performance of current diagnostic tools with criteria that include automated T-wave morphology analysis are needed. Investigations of this system’s ability to detect LQTS in pediatric patients and to identify less common LQTS subtypes are also warranted.
The authors have reported that they have no relationships relevant to the contents of this paper to disclose. Drs. Porta-Sanchez and Spillane contributed equally to this work.
All authors attest they are in compliance with human studies committees and animal welfare regulations of the authors’ institutions and Food and Drug Administration guidelines, including patient consent where appropriate. For more information, visit the JACC: Clinical Electrophysiology author instructions page.
- Abbreviations and Acronyms
- long QT syndrome
- morphology combination score
- principal component analysis
- principal component analysis-ratio 2
- receiver-operating characteristic
- Received July 5, 2016.
- Revision received October 18, 2016.
- Accepted October 20, 2016.
- 2017 American College of Cardiology Foundation
- Vyas H.,
- Hejlik J.,
- Ackerman M.J.
- Lehmann M.H.,
- Suzuki F.,
- Fromm B.S.,
- et al.
- Benhorin J.,
- Merri M.,
- Alberti M.,
- et al.
- Malfatto G.,
- Beria G.,
- Sala S.,
- Bonazzi O.,
- Schwartz P.J.
- Moss A.J.,
- Zareba W.,
- Benhorin J.,
- et al.
- Abu-Zeitone A.,
- Peterson D.R.,
- Polonsky B.,
- McNitt S.,
- Moss A.J.
- Etheridge S.P.,
- Compton S.J.,
- Tristani-Firouzi M.,
- Mason J.W.
- ↵Andersen MP, Xue JQ, Graff C, et al. A robust method for quantification of IKr-related T-wave morphology abnormalities. Presented at: Computers in Cardiology Conference; September 30−October 3, 2007, Durham, NC.
- Zhang L.,
- Timothy K.W.,
- Vincent G.M.,
- et al.
- Priori S.G.,
- Mortara D.W.,
- Napolitano C.,
- et al.
- Vicente J.,
- Johannesen L.,
- Mason J.W.,
- et al.
- Topilski I.,
- Rogowski O.,
- Rosso R.,
- et al.
- Vincent G.M.,
- Timothy K.,
- Fox J.,
- Zhang L.
- Drew B.J.,
- Ackerman M.J.,
- Funk M.,
- et al.
- Taggart N.W.,
- Haglund C.M.,
- Tester D.J.,
- Ackerman M.J.