Using QRS loop descriptors to characterize the risk of sudden cardiac death in patients with structurally normal hearts

The predictive value of non-invasive electrocardiographic examination findings for the risk of sudden cardiac death (SCD) in populations with structurally normal hearts remains unclear. This study aimed to investigate the characteristics of the QRS vectorcardiography of surface electrocardiography in patients with structurally normal hearts who experienced SCD. We consecutively enrolled patients who underwent vectorcardiography between March 2017 and December 2018 in a tertiary referral medical center. These patients didn’t have structural heart diseases, histories of congestive heart failure, or reduced ejection fraction, and they were classified into SCD (with aborted SCD history and cerebral performance category score of 1) and control groups (with an intervention for atrioventricular node reentrant tachycardia and without SCD history). A total of 162 patients (mean age, 54.3±18.1 years; men, 75.9%), including 59 in the SCD group and 103 in the control group, underwent propensity analysis. The baseline demographic variables, underlying diseases, QRS loop descriptors (the percentage of the loop area, loop dispersion, and inter-lead QRS dispersion), and other electrocardiographic parameters were compared between the two groups. In the univariate and multivariate analyses, a smaller percentage of the loop area (odds ratio, 0.0003; 95% confidence interval, 0.00–0.02; p<0.001), more significant V4-5 dispersion (odds ratio, 1.04; 95% confidence interval, 1.02–1.07; p = 0.002), and longer QRS duration (odds ratio, 1.05; 95% confidence interval, 1.00–1.10; p = 0.04) were associated with SCD. In conclusion, the QRS loop descriptors of surface electrocardiography could be used as non-invasive markers to identify patients experiencing aborted SCD from a healthy population. A decreased percentage of loop area and elevated V4-5 QRS dispersion values assessed using vectorcardiography were associated with an increased risk of SCD in patients with structurally normal hearts.


Introduction
Sudden cardiac death (SCD) is a significant public health problem worldwide, accounting for approximately 12-20% of all deaths annually [1][2][3]. To avoid such tragedy, it is important to identify high-risk patients vulnerable to cardiac arrest. However, developing a good tool is exceptionally challenging because detailed information on these patients is usually limited [4]. Surveillance of these survivors might be another method to help gradually characterize highrisk patients. Echocardiography and electrocardiography (ECG) have been widely validated as classifiers for certain populations to detect low ejection fraction and structural abnormalities in high-risk patients. In a healthy population without structural heart disease screened by echocardiography or ECG, some anomalies might exist but are difficult to identify, especially electrical abnormalities susceptible to malignant arrhythmias. Although ventricular arrhythmias (VAs) were not frequently detected, a previous study has proposed that the majority of SCDs could probably result from malignant arrhythmias [5]. VAs and non-VAs were considered to be in the same pathophysiology cascade but occur at different time points. Malignant arrhythmias probably have already degenerated to asystole, or sinus rhythm has already been restored when there is no detectable VA in aborted SCD. Thus, identifying the abnormal substrates of the ventricle to define high-risk groups has become an important issue because these subtle discontinuities may induce VA leading to SCD [6].
Although surface ECG is commonly used to evaluate the repolarization derangement of the QRS complex during the cardiac cycle, subtle QRS changes may not be detected by visual inspection, particularly in patients with structurally normal hearts. With the aid of a vectorcardiogram derived from ECG, a tiny repolarization abnormality could be identified.
We hypothesized that the vectorcardiogram of the QRS complex of surface ECG could be used to characterize the risk of SCD in patients with structurally normal hearts.

Study population
This retrospective study was performed in a tertiary referral medical center and was approved by the Institutional Review Board of Taipei Veterans General Hospital (IRB 2019-01-002C). All patients were aged 18 years or older. Oral and written informed consents were obtained from all participating adults or from the parents or legal guardians for incapacitated adults. The records and personal information of the patients were anonymized and de-identified before analysis. Reporting of the study conforms to the Strengthening the Reporting of Observational Studies in Epidemiology statement and the broader Enhancing the Quality and Transparency of Health Research guidelines [7].
This study consecutively enrolled patients who had histories of aborted SCD or palpitation that is due to atrioventricular node reentrant tachycardia (AVNRT) in our clinics and underwent vectorcardiography between March 2017 and December 2018. These patients didn't have structural heart diseases (e.g., dyskinesia or hypertrophy of left ventricles), histories of congestive heart failure, or reduced ejection fraction (<50%) [8,9], and the attribution to non-cardiac causes wasn't favored in patients with a history of aborted SCD. A total of 315 patients were investigated. Patients who had a history of aborted SCD were classified into the case group; these patients had a cerebral performance category score of 1 at baseline. Conversely, patients with no history of SCD were classified into control group. SCD was defined as death from cardiac arrest occurring within 1 hour from symptom onset [10]. A shockable rhythm was presented as the initial rhythm or during resuscitation. Propensity score matching was performed to minimize the confounders. The cases and controls were matched at a 1:2 ratio using a 0.10 caliper for identical characteristics of age, sex, and histories of hypertension and coronary artery disease. Fifty-nine patients with a history of aborted SCD and one hundred three controls were included in the analysis (Fig 1).

Morphological complexity of the 12-Lead QRS wave
The 12-Lead ECG was taken at a stable stage using LabSystem™ PRO (Boston Scientific, Boston, MA, USA) with a 1-min duration for QRS descriptor analysis. The QRS descriptors, including the percentage of the loop area (PL), loop dispersion (LD), and inter-lead QRS During the study period, 315 patients were enrolled. The patients were then separated into two groups: with and without a history of aborted sudden cardiac death (SCD). Propensity analysis was performed to minimize the confounders, and the cases and controls were matched at a 1:2 ratio using a 0.10 caliper for identical characteristics of age, sex, and histories of hypertension and coronary artery disease. Ultimately, there were 59 patients with a history of aborted SCD and 103 control patients without a history of aborted SCD. (Abbreviations: Hx, history; LV, left ventricular; SCD, sudden cardiac death).
https://doi.org/10.1371/journal.pone.0263894.g001 dispersion (IQRSD) are measured using the ECG vector in the constructed space spanned by the eigenvector of 12 lead ECG matrix. All collected variables, including disease history, risk factors, and comorbidities, were obtained from primary or secondary hospital discharge diagnoses and outpatient visits, and the information was confirmed via telephone interviews. Targeted co-morbidities, including hyperlipidemia, CKD, old CVA, prior CAD, were determined by using the International Classification of Diseases (ICD) 9 codes from the medical record at the time of examination. In addition to the index aborted SCD event, the number of other major arrhythmic events was also recorded, including sustained ventricular tachycardia, ventricular fibrillation, appropriate implantable cardioverter-defibrillator shocks, aborted cardiac arrest, and SCD. The ECG and echocardiographic parameters, including the heart rate, PR interval, QRS duration, corrected QT(QTc) interval, and left ventricular ejection fraction (LVEF), were also compared between the two groups.

ECG decomposition via principal component analysis (PCA)
The energy of the QRS complex was analyzed. An ECG matrix, X 8xN , was used, and each row corresponded to standard ECG leads (I, II, V1, V2, V3, V4, V5, and V6). N represented the number of cases for each lead. Only 8 of the 12 leads were used because of algebraic interdependency to reconstruct an orthogonal-8-lead system via PCA in the signal space [11]. Nightynine percent of the ECG energy was calculated in the first three PCAs [12], and the vectors associated with the first three PCAs were incorporated to investigate and describe the spatiotemporal variations of the energy in the QRS complex. This method has been described in detail elsewhere [13].

QRS loop descriptors
The QRS loop in a two-dimensional plane was constructed by sampling the QRS complex of the first principal component as the x-coordinate and the corresponding data of the second principal component as the y-coordinate [12][13][14]. The minimum rectangle was created and accommodate the planar loop, and the rectangle was divided equally into N (N = 4900 in this study). PL was defined as the percentage of cells within the QRS loop, indicating the regularity of the loop. LD was defined as the total number of cells that were passed through by the loop [13,14]. IQRSD was based on the angular difference between the reconstructed adjacent vectors obtained from QRS loop decomposition via PCA [12,14], and it represented the dissimilarity of the shape between the QRS complexes of two ECG leads [13] (Fig 2, lower part), which could not be detected by visual inspection on surface ECG (Fig 2, upper part). IQRSDs and other parameters were compared between the two groups.

Diagnosis of SCD
All patients with a history of aborted SCD were diagnosed in accordance with the diagnostic criteria. Arrhythmogenic right ventricular cardiomyopathy (ARVC) was diagnosed in accordance with the 2010 modified Task-Force criteria [15]. Brugada syndrome was diagnosed on the basis of documented spontaneous or flecainide-induced ECG changes [16,17]. Long QT syndrome was diagnosed on the basis of the proposed diagnostic score [18]. An idiopathic etiology was considered when these abovementioned diseases were not identified.

Statistical analysis
All analyses were performed using the SPSS software (version 24.0, SPSS, Inc., Chicago, IL, USA) and R (version 3.6.3). The baseline characteristics of the patients were reported as means ± standard deviations for continuous variables and as percentages for categorical variables. The Chi-square test was performed to compare the categorical variables between the groups. The independent t-test was used to compare the continuous variables between the two groups, while the one-way ANOVA was used to compare the values between the three groups. Multivariate analysis was performed for variables with a p-value of <0.1 in the univariate analysis. Receiver operating characteristic (ROC) curve analysis was used to assess the values of the continuous variables for risk stratification, and optimal cutoff values were calculated sequentially according to the specificity and sensitivity. Statistical significance was set at pvalue of <0.05.
Based on the findings shown in Fig 3A, there was no subject in the control group with a V 4-5 dispersion value of >60˚. To shed some light on the features of patients at risk for SCD, we demonstrated the baseline characteristics and ECG and echocardiographic findings of the patients with V 4-5 dispersion values of >60˚in S1 Table. We noticed that most of these patients were men (73.3%), and the mean age was 51 years. Histories of hypertension and coronary artery disease accounted for 33.3% of all cases. Regarding the etiologies of SCD, all the patients with a history of aborted SCD have received routine coronary angiography, and none of them have acute coronary syndrome. One case was Brugada syndrome (6.7%); one was ARVC (6.7%); and 13 were idiopathic (86.7%). In addition, there were no significant abnormalities in the ECG and echocardiographic parameters.

Multivariate analysis
The odds ratios (ORs) for SCD derived from the multivariate analysis are shown in Table 3. In model 1, the patients with a longer QRS duration (OR, 1.05; 95% confidence interval (CI), 1.00-1.10; p = 0.04) had a higher risk for SCD. In model 2, the patients with a smaller PL (OR, 0.0003; 95% CI, 0.00-0.02; p<0.01) and larger V 4-5 dispersion (OR, 1.04; 95% CI, 1.02-1.07; p<0.01) had a higher risk for SCD. Moreover, we used ROC curve analysis to assess the values of the PL, V 4-5 dispersion, and QRS duration for predicting SCD or non-SCD (S1 Fig In the SCD group, 28.9% of the patients had two SCD events, and 72.1% had only one event. To investigate the PL, V 4-5 dispersion, and QRS duration values in relation to the SCD events, we further compared these parameters with regard to the SCD event numbers, and the results are shown in Fig 4. The PL of the patients with one event (57.4±13.3%) and two events (54.1 ±14.6%) was lower than that of the controls (66.0±8.8%) (p<0.01). The patients with one event (47.6±18.5˚) and two events (41.7±19.5˚) had higher V 4-5 dispersion values than the controls (28.9±12.8˚) (p<0.01). The QRS duration of the patients with one event (95.1±10.8 ms) was longer than that of the controls (89.8±9.9 ms). However, there was no significant difference in the QRS duration between the patients with two events (90.8±9.4 ms) and controls (p = 0.10).
In addition, the baseline demographic characteristics and studied parameters were also compared between the patients with and without shockable rhythms at initial presentation (S3 Table). There was no difference in the underlying diseases, vectorcardiographic parameters, and most ECG and echocardiographic parameters between the two groups. Only the PR interval was significantly longer in the shockable rhythm group than in the other group, which was probably attributed to the effect of multiple factors, including the heart rate.

Main findings
In this study, surface ECG was used to create a vectorcardiogram, which could define the difference between patients with symptomatic AVNRT and patients with a history of aborted SCD. The main findings of this study are as follows: (1) In the patients with structurally normal hearts, increased QRS duration on ECG, decreased PL, and elevated V 4-5 QRS dispersion values assessed on vectorcardiography were associated with an increased risk of SCD. These parameters were also associated with SCD independently of the presence of other ECG abnormalities in a multivariate model. Of note, approximately 20% of the SCD cases had V 4-5 dispersion >60˚, whereas none of the controls had V4-5 dispersion >60˚. (2) the PL and V 4-5 QRS dispersion values were correlated with the number of SCD events.

ECG findings in relation to SCD risk stratification
The majority of SCD cases occurred in the adults above 35 years of age, and nearly 50% of the events were attributed to VAs, although the actual incidence of VA in patients with a history of aborted SCD remains unknown because it inevitably progresses to unwitnessed asystole [5]. The predictive value of repolarization abnormalities in SCD has been evaluated by measuring the R-wave and T-wave dispersion in previous studies [19,20]. One of the models was constructed on the basis of a 12-lead ECG to calculate the T-wave area dispersion of leads I, II, and V4 through V6. Patients with lower values are more vulnerable to SCD, which suggests that repolarization heterogeneity is implicated in SCD [20]. However, subtle abnormalities in depolarization are sometimes difficult to recognize on ECG only, especially in patients with structurally normal hearts. With the aid of vectorcardiograms derived from ECG, we can characterize the depolarization abnormalities in patients with a history of aborted SCD by investigating the energy of the QRS complex. We observed that decreased PL and elevated V 4-5 QRS dispersion values assessed on vectorcardiography were associated with an increased risk of SCD in the patients with structurally normal hearts. Further studies are warranted to validate these observations in the future.

Role of the QRS loop descriptors in SCD
To our knowledge, this is the first time that QRS loop descriptors have been used to characterize aborted SCD. Our analysis showed that the patients with a smaller PL and more significant V 4-5 dispersion were at risk for SCD. Moreover, a longer QRS duration was also related to the risk for SCD [21].
The PL indicates the regularity of the loop, and a smaller PL could result from asynchrony in the myocardial tissue and influence IQRSD [13]. Moreover, IQRSD refers to the similarity between adjacent leads. Hence, a highly heterogeneous substrate could result in a higher IQRSD value. Thus, it is reasonable that the PL of the patients with a history of SCD was small, and the IQRSD value was higher in the SCD group than in the control group in our study ( Table 2).
The findings of the vectorcardiography analysis can be discussed from different perspectives. First, a previous study has proposed that QRS dispersion is related to the electrophysiological conduction of the myocardium; thus, more significant dispersion will lead to more VAs [22]. In our study, the SCD group had a lower regularity of the QRS loop descriptors than the control group. The irregularity property of the QRS loop descriptors represents the non-identical conductive features of the myocardium. It could introduce arrhythmogenic effects and even induce SCD.
Second, the concept of failed activation was proposed on the basis of a current-to-load mismatch in the depolarization phase. In studies of explanted human hearts and experimental mouse hearts, a depolarization disorder might result in VA or SCD [23]. Herein, tiny QRS abnormalities may not be detected on a 12-lead surface ECG (Fig 2). We used the QRS loop descriptors to substantially reveal the difference in the QRS complex between the two adjacent leads in a signal space and to show the main inconsistency and failed activation [12,14]. In our study, the inter-lead dispersion of QRS might represent the dissimilarity of the energy of the QRS complexes from the two leads, and this is the angle between the two reconstructed vectors [13]. The more dissimilar they are, the large the angle will be. We found that V 4-5 had significant dispersion in the SCD group compared with the control group. The significantly different dispersion within the precordial region might also indicate the main inconsistent activation of the precordial leads.
Finally, SCD primarily results from left ventricular disease [24]. However, in populations with structurally normal hearts, this risk is considerably difficult to identify. In our study, we used a high value of inter-lead dispersion to identify left ventricle disorder in the patients with a history of aborted SCD. Although this tool could not accurately identify the etiologies, it could still be a promising approach to point out the lethally subtle changes in cardiac repolarization.

Role of the QRS duration in SCD
It has been proposed that a wide QRS duration might contribute to SCD [25]. In our study, we found that the QRS duration in the SCD group was relatively longer than that in the control group (Tables 2 and 3), even though there were no structural heart disorders in these patients. Hence, for the general population with structurally normal hearts and longer QRS duration, potentially lethal events should still be considered.

Proportion of initial shockable rhythms in survivors
The proportion of VA (42.4%) among the survivors in this study was relatively low. According to the pooled dataset from the COSTA group, the percentage of patients presenting with VA among European patients who experienced out-of-hospital cardiac arrest (OHCA) has already reached 37-38% [26]. However, the differences between continents and ethnicities in these data probably need to be considered. A lower detection rate of a shockable rhythm at the initial presentation in Asia has been reported by Berdowski et al. [27]. In addition, real-world data from the THUNDER study based on the Taiwanese population also disclosed that the incidence of a shockable rhythm in patients who experienced OHCA was only 7.9% [3]. According to this registry, 17.8% of the patients with an initial shockable rhythm and 2.7% of those with an initial non-shockable rhythm would be able to survive to discharge, indicating that only 36% of all survivors who experienced OHCA had shockable rhythms at the initial presentation. Taken together, our results are similar to the real-world Taiwanese registry data and was consistent with other study data in relation to the difference among continents.

Limitations
Our study has several limitations. First, there is a possibility of selection bias and survivorship bias because only patients with a cerebral performance category score equal to 1 were enrolled. We had no access to the patients who didn't survive the OHCA as well as the severely ill patients who were unable to undergo 1-min vectorcardiography owing to poor clinical status; therefore, the findings of our study were probably related to the survival of cardiac arrest instead of the risk of the cardiac arrest itself. In addition, the ECG parameters were collected and elaborated after the event; thus, the differences found in our study might also have a chance to be a result of these events. Second, the etiologies of SCD in the patients were heterogeneous. Forty-one patients had an idiopathic etiology; these patients need further evaluation in the future. Third, our hospital is a tertiary referral medical center in the Taipei metropolitan area, which encompasses a population of 6.6 million. Unfortunately, we did not have a detailed number of aborted SCD events in this area; thus, we could not precisely assess the proportion of patients included in this study. However, an estimation can be made according to a previous study conducted in central Taiwan with a population of 2.7 million [3]. In that study, there were 1629 patients who experienced OHCA in a year, of whom 3.9% survived. Hence, at a rough estimate, there were 285 patients with a history of aborted SCD in this region during the study period, of whom 25.3% were referred to our hospital. Fourth, we had limited access to the results of cardiac magnetic resonance imaging, genetic evaluation, and brain computed tomography angiography that might be routine examinations following SCD, especially when patients were referred to our clinic. Furthermore, the surface ECG parameters were recorded using LabSystem™ PRO (Boston Scientific, Boston, MA, USA). More studies are warranted for the application to more systems in the future.

Conclusions
The QRS loop descriptors and vectorcardiographic parameters derived from surface ECG could be used as non-invasive markers to identify patients at a high risk for SCD. Specifically, low PL and high V 4-5 QRS dispersion values are associated with patients who had structurally normal hearts and experienced aborted SCD. We performed a subgroup analysis to determine the risk factors in the two groups. We found that sex, age, and creatinine level did not significantly contribute to the incidence of SCD. Instead, we observed that the patients with a relatively low LVEF had a tendency to have a history of aborted SCD, although this was not significant. (Abbreviations: Cr, creatinine; LVEF, left ventricular ejection fraction; OR, odds ratio; QTc interval, corrected QT interval; SCD, sudden cardiac death). (TIF) S1 Table. Characteristics of the patients with a V4-5 dispersion value of >60˚. We used a cut-off value of >60˚for V4-5 dispersion to characterize patients at risk for sudden cardiac death. (DOCX) S2 Table. ROC analysis of the parameters for predicting sudden death cardiac death(SCD) or non-SCD. Receiver operating characteristic curve analysis was used to assess the values of the continuous variables for risk stratification, and optimal cutoff values were calculated sequentially according to the specificity and sensitivity. (DOCX) S3 Table. Baseline characteristics of the patients with a history of aborted sudden cardiac death with shockable and non-shockable rhythms. There was no difference in the underlying diseases, vectorcardiographic parameters, and most ECG and echocardiographic parameters between the two groups. Only the PR interval was significantly longer in the shockable rhythm group than in the other group. (DOCX)