Ventricular Electrical Delay Measured From Body Surface ECGs Is Associated With Cardiac Resynchronization Therapy Response in Left Bundle Branch Block Patients From the MADIT-CRT Trial (Multicenter Automatic Defibrillator Implantation-Cardiac Resynchronization Therapy)

Introduction

See Editorial by Killu and Cha

Current results from major clinical trials investigating the optimized management of patients with systolic heart failure (HF) indicate that cardiac resynchronization therapy (CRT) is a treatment of choice that demonstrates a significant benefit for most patients. The MADIT-CRT study (Multicenter Automatic Defibrillator Implantation-Cardiac Resynchronization Therapy) showed that CRT-D therapy was associated with a 34% reduction to the risk of an HF event or death in comparison with implantable cardioverter defibrillator therapy (ICD). When considering patients with a QRS prolongation ≥150 ms, CRT seems to be associated with the greatest benefit. Nevertheless, the role of QRS duration (QRSd) in the prognosis of patients with implanted CRT devices is under great scrutiny and remains controversial¹; whereas left bundle branch QRS morphology seems to be the most important in identifying patients who benefit from CRT.^2,3 However, 20% to 30% of HF patients presenting with QRSd >120 ms, left bundle branch block (LBBB) and left ventricular ejection fraction (LVEF) <35% fail to respond to prescribed CRT therapy (nonresponders).^4–6 Therefore, it is of paramount importance to improve our ability to predict CRT outcome before implantation to maximize the cost benefit of CRT.

It is generally accepted that CRT has a beneficial effect in patients with large left ventricle dyssynchrony⁷ as manifested by LBBB morphology in an ECG. Unfortunately, QRS morphology does not directly reflect the electrical dyssynchrony. Technology for the accurate measurement of electrical dyssynchrony from high-frequency ECGs was recently introduced⁸ and showed that averaged high-frequency components in a QRS complex region can provide information about the temporal-spatial distribution of depolarization. The comparison of depolarization activation patterns in different ventricular segments (leads) and computation of depolarization interlead delays was interpreted as electrical dyssynchrony (measured in milliseconds). In this article, we validate the high-frequency ECG approach and introduce a new parameter for measuring ventricular electrical delay (VED) from high-resolution 12-lead ECG recordings. We studied the associations between baseline VED values and MADIT-CRT combined end points and investigated whether VED measurements could help identify patients benefitting from their CRT device.

Methods

Data, analytic methods, and study materials will not be made available to other researchers for the purposes of reproducing the results or replicating the procedure.

Study Population

MADIT-CRT was designed to determine whether CRT-D therapy would reduce the risk of death or HF in patients with mild cardiac symptoms, a reduced ejection fraction, and a wide QRS complex. A total of 1820 patients were enrolled in the study, consisting of 1046 ischemic cardiomyopathy patients with New York Heart Association I or II and 774 nonischemic cardiomyopathy patients with New York Heart Association II, LVEF ≤30%, and QRSd >130 ms. Patients provided informed consent and were randomized in a 2:3 ratio to receive ICD or CRT-D therapy. The multicentric MADIT-CRT trial was approved by multiple institutional review boards at each involved center. The detailed design of the MADIT-CRT trial, including device programming and interrogation, have previously been described.^9,10

Only patients from the CRT-D arm had a digital ECG acquired before implantation (baseline). Hence, we analyzed 1089 recordings to predict the MADIT-CRT primary combined end points, that is, HF event or death. We also investigated the association between baseline VED values and 12-month changes in left ventricular end-systolic volume (LVESV), left ventricular end-diastolic volume (LVEDV), left atrial volume, LVEF, and LV dyssynchrony. For the purposes of comparison, we also used survival data from 520 LBBB patients in the ICD arm.

Clinical ECG and Echocardiographic Data

Body surface ECGs were acquired using an H12+ Mortara recorder (Mortara Instrument Inc, Milwaukee, WI) delivering digital 12-lead ECG signals sampled at 1 KHz and an amplitude resolution of 3.75 microvolts. The lead configuration used for the recordings was the Mason-Likar system. The signal from the first 10 minutes of the ECG recording, acquired in the resting supine position, was used in the study.

Echocardiographic data were available for 488 CRT-D patients both at enrollment and at 1-year follow-up. Analysis of echocardiographic images provided the LVEDV, LVESV, and left atrial volume. LVEF values were obtained by angiographic or echocardiographic methods.⁹ The volumes were estimated by averaging from 2 and 4 chamber images according to Simpson method.¹¹

Computation of VED

VED computation was based on the high-frequency ECG approach^8,12 and was optimized to 1-kHz 12-lead ECG Holter recordings. Only the QRS complexes from sinus beats were used to compute VED. First, the ECG signals from leads V₁, V₂, V₅, and V6 (Figure 1A, left) were digitally filtered using 3 frequency bands (F₁, 150–250 Hz; F₂, 250–350 Hz; and F₃, 150–350 Hz), leading to a set of 12 enveloping signals (F_1–3 for 4 leads). Next, the median high-frequency QRS (HFQRS) shape was computed (Figure 1A, right) from each enveloping signal leading to 12 HFQRS. In each HFQRS, we detected the time location of the amplitude maxima. Finally, we measured the longest time difference between maxima in the lateral and septal V-leads for each frequency band. This process was repeated using the center of masses of HFQRS instead of the maximum amplitude and used the average location between centers of mass and amplitude maxima. The VED value was the average value across the 9 interval delays (3 for maxima, 3 for centers of mass, and 3 for average locations between centers of mass and maxima). VED values are expected to be positive in LBBB patients and negative in patients with right bundle branch block (RBBB). Indeed, in RBBB patients the late activation of the right ventricle means that the envelope maxima (and centers of masses) in right leads (V₁ and V₂) are located after the ones in lateral leads (V₅ and V₆). The exact computation procedure also included annotation jitter removal and baseline correction.

• Download figure
• Download PowerPoint

We illustrate the importance of the VED method in Figure 1B and 1C in which we plotted 2 ECGs from 2 LBBB patients with a QRSd of 130 ms (Figure 1B) and 138 ms (Figure 1C). We display the median HFQRS and the associated VED values on the bottom panels. Figure 1B presents an example of a patient who responded strongly to CRT with a 12-month change of LVESV of −34.1%. The bottom panel shows the envelope signals and the large delay between upper and lower precordial leads leading to a VED=46.4 ms, despite a QRSd=130 ms. Figure 1C shows the tracings for a patient with poor CRT response, that is, LVESV changes equal to −6.3%. Despite the slightly higher QRSd (138 ms), the ECG of this subject reveals a very different VED value, that is, a different configuration of the high-frequency envelopes of the QRS signal.

Statistical Analysis

We present a retrospective analysis of the MADIT-CRT data using an innovative computerized ECG method. The first quartile of VED values was used as a threshold for comparison of the baseline clinical characteristics and for further statistical analyses. Continuous variables were compared using the Wilcoxon rank-sum test and are presented as means and SDs; categorical data were computed with a χ² test and presented as counts and percentages.

The cumulative probability of the MADIT-CRT primary end point (HF or all-cause mortality) was displayed using Kaplan–Meier plots.¹³ The significance of the cumulative risk of the end point was estimated using the log-rank test. A multivariate Cox proportional hazards regression model¹⁴ adjusted for creatinine, sex, LVESV index (by body surface area index), prior hospitalization for chronic HF, QRSd, and systolic blood pressure ≥140 mm Hg was used for estimation of the association of VED with the risk of primary end points. The Akaike information criterion (AIC) was estimated to compare the relative performances of the multivariate Cox proportional hazards models. Uno’s concordance statistic, essentially a type of C statistic measure for censored survival data, was used to estimate the predictive accuracy of the models.¹⁵

We also examined the predictive value of VED in prespecified MADIT-CRT subgroups. P values <0.05 were considered significant. Statistical analysis was prepared using SAS software (version 9.4, SAS Institute, Cary, NC).

Results

Study Population

Figure 2 provides a consort chart describing the subsets of the MADIT-CRT population analyzed in this study. Focusing on the subjects with CRT-D, a subset of the trial including 1089 individuals was used. One hundred four of these subjects were not analyzed because of missing ECG recordings, and 32 subjects were removed because of an insufficient number of usable sinus QRS annotations (required N≥120 in the first 10 minutes of recording). Insufficient ECG signal quality was the primary cause of the rejection of these recordings. In addition, 2 patients had Holter recordings but were enrolled in the ICD arm of the study, and 2 other patients did not have an ECG conduction disturbance designation. These 4 recordings were excluded from the analysis for this reason. Ultimately, these patients formed the LBBB cohort (N=676), RBBB cohort (N=113), and intraventricular conduction delay (IVCD) cohort (N=160).² In LBBB patients, we had echocardiographic measurements in a set of 488 subjects for whom hemodynamic parameters were available pre- and postimplantation.

• Download figure
• Download PowerPoint

Higher VED Is Associated With Better CRT Outcome

VED was computed in LBBB patients (N=676; median, 49.1 ms; interquartile range, 33.6 ms) at baseline and dichotomized at the first VED quartile (Q1=31.2 ms) because low VED indicates lesser electrical dyssynchrony. Comparison of the baseline clinical characteristics is reported in Table 1. Comparison of the probability of HF events or death using Kaplan–Meier analysis in LBBB groups dichotomized using the VED Q1 value is shown in Figure 3A. The LBBB patients with VED <Q1 at baseline were at higher risk (35%) than patients with VED ≥Q1 (14%, P<0.001). Moreover, Figures 3B and 3C present the Kaplan–Meier plots by VED values above and below its first quartile in patients with QRSd <150 ms and QRSd ≥150 ms.

• Download figure
• Download PowerPoint

Multivariate Cox regression models for hazard ratios (HRs) using VED in both continuous and dichotomized forms are presented in Table 2. Table 2 shows a model for HR using the dichotomized VED value (HR, 2.34; 95% confidence interval, 1.53–3.57; P<0.001) adjusted for QRSd with an AIC of 1345 and C statistics equal to 0.711. The model adjusted for the same covariates except QRSd shows an AIC of 1348 and C statistics of 0.706 (VED HR, 2.83; 95% confidence interval, 1.92–4.17; P<0.001). In comparison, a model with QRSd but without VED has an AIC of 1358 and a C statistics of 0.689. AIC value of dichotomized model adjusted for QRSd performs only slightly better but also shows that VED contributes more to the prediction of primary end points than QRSd. The model for continuous VED values (Table 2) showed that each 10 ms increase in VED value is associated with a 13% decrease for the risk of MADIT-CRT end point (P=0.003). Additional information for these models is available in the Data Supplement. Moreover, separate multivariate Cox regression models in ischemic and nonischemic patients showed that a 10 ms increase in VED value is associated with a 10% (P=0.067) and 17% (P=0.011) decrease of end point risk, respectively.

VED Is Associated With Hemodynamic Changes and Ventricular Arrhythmias

For LBBB patients, we report changes at 12 months after CRT implantation in LVEDV, LVESV, left atrial volume, LVEF, and LV dyssynchrony measurements. These results are summarized in Figure 4 for patients with VED values above (or equal to) and below 31.2 ms (Q1) at the baseline. Comparing the groups of patients with low baseline VED values (<Q1) to the rest of the cohort (≥Q1), we found that CRT-induced changes in LVEDV (−17.9±9.5% versus −24.0±11.8%, P<0.001), LVESV (−28.8±13.1% versus −36.8±14.9%, P<0.001), LVEF (9.8±4.5% versus 12.5±5.2%, P<0.001), and left atrial volume (−26.1±11.3% versus −30.5±12.0%, P<0.001) were higher in those patients with VED ≥Q1 at baseline. In addition, the beneficial changes in LV dyssynchrony acquired from echo speckle tracking were larger in those patients with VED ≥Q1 (−31±92% versus −58±78%, P=0.015).

• Download figure
• Download PowerPoint

In examining the occurrence of ventricular arrhythmias (VT/VF) in the LBBB cohort in relation to VED values, we found a larger percentage of patients with VT/VF in those with low baseline VED values (VED<Q1) than in the rest of the group: 24% versus 16%, P=0.031. This observation may also suggest that higher VED before CRT implantation could be associated with better patient outcome.

VED Values and Their Association With Outcome in Subgroups

We provide a forest plot in Figure 5 describing the HR in LBBB patients dichotomized at Q1 of VED. We used the model adjusted for QRSd (Table 2). The top line represents the HR for the entire LBBB population and shows that patients with a baseline VED <Q1 have a 2.34× higher risk of the study combined end points in comparison with the group with baseline VED ≥Q1. HR in subgroups defined using clinical baseline characteristics such as age, gender, and other relevant clinical variables were similar. Hence, we report that there is no significant interaction between these variables and the prognostic value of VED for predicting the primary end point of the trial. It is worth emphasizing that HRs for subgroups by QRSd at 150 ms show nearly similar HRs (2.78, 95% confidence interval, 1.32–5.83; P=0.007 versus HR, 2.11; 95% confidence interval, 1.21–3.65; P=0.008). These results suggest that VED may be useful as a predictor of HF events or death in patients with QRSd both lower and higher than 150 ms.

• Download figure
• Download PowerPoint

We investigated the association between the QRSd and VED values using the Pearson correlation. The result demonstrated a significant but weak correlation: R=0.50 (P<0.001).

VED Values in the Non-LBBB Population

Despite the lack of response to CRT in non-LBBB patients described in the MADIT-CRT study, we investigated the values of VED in patients with other ventricular conduction abnormalities, that is, RBBB and other intraventricular conduction defects (IVCD). The median VED value was −26.5 ms (interquartile range, 27.2 ms) for RBBB patients and 16.1 ms (interquartile range, 19.1 ms) for IVCD patients, respectively. Both cohorts were dichotomized using specific VED values because their values are very different from the group of LBBB patients. The risk of end point was 15% at 4 years for VED ≥Q3 (Q3=−10.9 ms) in RBBB patients and 29% for the other group (VED <Q3). This difference was not significant (P=0.14). In IVCD patients, the risk was equal to 20% and 44% for VED <7.3 ms (Q1 in the IVCD population) and VED ≥7.3 ms, respectively (P=0.05).

Discussion

In this retrospective study, we present a new parameter called VED that measures the delay between the primary components of the filtered QRS measured in the septal and lateral leads of resting ECGs. We studied the association between VED and the risk of HF or all-cause mortality in MADIT-CRT patients. We demonstrated that the MADIT-CRT LBBB patients with low VED before CRT implantation are at higher risk of study combined end points and have worse CRT outcomes. The electrical septal-lateral delay in LBBB patients measured by VED is a type of electrical dyssynchrony that CRT pacing should correct.⁸ Our results may support the use of VED for optimal selection of patients benefiting from CRT and risk stratification for combined end points.

Moreover, the risk associated with low preimplantation VED values does not seem to be modulated by the clinical factors shown in Figure 5. Among others, this figure also presents QRSd, one of the criteria in the CRT guidelines.^16,17 Although VED seems to derive different information than QRSd, all time delays are examined inside the QRS complex and, therefore, VED cannot be longer than QRSd. As expected, patients with VED <Q1 have a narrower QRS complex (149.8 ms) than patients with VED ≥Q1 (166.4 ms). But, as shown in Figure 6, VED and QRSd quartile stratifications lead to a different level of association with CRT benefit which is stronger in the case of VED.

• Download figure
• Download PowerPoint

Although VED seems to be more strongly predictive than QRSd of CRT benefit, we observed that the combination of both measurements leads to a better HR model. As the analysis of both the AIC and Uno concordance statistics shows, the inclusion of QRSd only slightly increases model prediction. This applies to both dichotomized and continuous functional forms. It is also worth mentioning that VED maintained its important predictive value even after QRSd was added to the model (Table 2; Expanded details in the Data Supplement). In comparing multivariate models which use QRSd and VED separately, the AIC and C statistics are better for models with VED when compared with those with QRSd. This again applies to both continuous and dichotomized functional forms of both variables. Other clinical parameters such as age, ischemia, LVEF, LVEDV, and LVESV did not show interaction with VED and emphasized its complementarity to these relevant clinical markers of CRT response. QRSd correlated with VED, but VED provided independent prediction of LBBB response in Cox models.

One could speculate that the VED value might be impacted by the presence of myocardial scarring. Therefore, we also investigated the differences in VED values between ischemic and nonischemic groups of patients. In Figure 5, we showed that the subgroup of nonischemic patients has a similar HR to the subgroup of ischemic patients.

In general, CRT outcome in HF patients with non-LBBB ECG morphology is less successful than in LBBB patients.^11,16–22 In our study, we also evaluated the VED parameter in 113 RBBB and 160 IVCD patients (limited-size subgroups). In both these groups, lower absolute VED values (negative in the case of RBBB) were associated with a lower risk of the combined end point, although nonsignificant in the case of RBBB patients. This opposite behavior in IVCD patients may be explained by the fact that VED values are very low (both ventricles work relatively synchronously). Therefore, the resynchronization effect is limited. And if there is limited or no effect of CRT, the patient with the lower absolute VED (ie, the patient with synchronous ventricle contractions) should be exposed to a lower risk of end point than the patient with higher VED (ie, the patient with asynchronous ventricle contractions). This may also apply to the RBBB population. Such behavior may suggest that CRT in non-LBBB patients has a limited effect which corresponds to the European Society of Cardiology guidelines defining stronger restrictions for CRT in non-LBBB patients than in LBBB patients. It is also apparent that the use of VED in the non-LBBB population should be investigated in future studies.

Although VED association with CRT outcome in specific cases remains unclear, the VED has a strong association with the risk of end point in LBBB patients. Specifically, a low VED points to LBBB patients with lower CRT benefit. Subsequently, the risk of end points in those patients with lower VED may be compared with the risk of end points in patients without CRT, as presented in Figures 3 and 6. This shows that the risk of combined end points in CRT patients with low VED (<Q1) is very close to the end point risk in patients with ICD only (520 LBBB patients). Moreover, Figure 3B and 3C show that VED provides new information in addition to QRSd which may be important when considering borderline cases for CRT. Therefore, we may state that VED has potential to be useful in identifying patients who might benefit from CRT therapy.

Limitations

VED should be used with the LBBB population only; results connected to the RBBB and IVCD population led to an opposite association with a MADIT-CRT end point, although nonsignificantly in the case of RBBB.

Conclusions

In this retrospective study, we showed that VED is associated with CRT response in LBBB patients and may provide new information in addition to existing clinical markers. We also showed that LBBB CRT recipients with a low VED value show no or limited benefit from the therapy in comparison to those with high VED. We think that this noninvasive computerized ECG method may assist physicians in the optimal selection of patients for CRT therapy.

Footnotes