Refining cardiac resynchronization therapy selection in borderline QRS duration patients

Abstract

Background

Cardiac resynchronization therapy (CRT) improves outcomes in heart failure patients with left bundle branch block (LBBB) and QRS of ≥150 ms, but benefits with borderline QRS (120–149 ms) remain uncertain. Electrical (ED) and mechanical dyssynchrony (MD) may refine patient selection, although reliable predictors are lacking.

Objective

To examine clinical predictors of dyssynchrony in patients with borderline QRS duration (QRSd) undergoing CRT.

Methods

We retrospectively studied 336 CRT recipients (2006–2015). Vectorcardiograms quantified QRS area (QRSa), with ED defined as QRSa of ≥95 μVs. MD was assessed via speckle-tracking echocardiography. Logistic regression identified predictors of dyssynchrony. QRSa indexed to left ventricular end-diastolic volume (QRSa/LVEDV) was evaluated in logistic and Cox models, stratified by a QRSd of 150 ms threshold.

Results

ED and MD were present in 53.0% and 74.4% of patients. Shorter height was independently associated with ED, while female sex predicted MD. Among patients with QRSd of <150 ms, ED and MD were less prevalent, and height was the sole independent predictor of ED. Over a median follow-up of 28.3 months, 241 patients (71.7%) experienced a composite endpoint, with similar event rates between QRS subgroups. QRSa/LVEDV was the strongest predictor of MD (odds ratio: 26.83; P = .0004) and reduced risk of adverse outcomes (hazard ratio: 0.172; P = .0006) in the <150-ms subgroup, outperforming QRSd and sex. When MD was added, the QRSa/LVEDV association attenuated (hazard ratio: 0.35; P = .07).

Conclusion

Among LBBB patients with borderline QRSd, QRSa/LVEDV was more strongly associated with MD and outcomes than QRSd or other factors, supporting its potential role in refining patient selection.

Graphical abstract

Key Findings.

• ▪QRS area/left ventricular end-diastolic volume (QRSa/LVEDV) strongly predicted the composite endpoint among left bundle branch block patients with QRS duration (QRSd) of <150 milliseconds, outperforming QRSd, sex, and height.
• ▪Adding mechanical dyssynchrony attenuated the QRSa/LVEDV association, indicating it predicts outcomes largely by identifying patients with mechanical dyssynchrony.
• ▪Because no non-cardiac resynchronization therapy (CRT) comparator was studied, QRSa/LVEDV could not be established as a marker of CRT-specific benefit, and prospective multicenter validation is needed.

Introduction

Cardiac resynchronization therapy (CRT) is an established treatment for heart failure (HF) patients with left bundle branch block (LBBB) and reduced left ventricular ejection fraction (LVEF). In patients with LBBB and LVEF of ≤35%, current guidelines recommend stratification based on QRS duration (QRSd).¹ Specifically, CRT with biventricular pacing carries a class I indication when QRSd is ≥150 ms. However, the use of CRT in patients with a QRSd between 120 and 149 ms remains controversial. European guidelines consider CRT implantation among patients with LBBB and a QRSd between 120 and 130 ms a class III indication, but allowable when the QRSd is 130–149 ms.² While recent HRS pacing guidelines allow for CRT among patients with a QRSd between 120 and 149 ms, this document recommends incorporating additional clinical factors, such as sex and height, into CRT decision-making for patients with borderline QRSd.³ However, it remains unclear which patient characteristics are the most predictive of benefit in this group.

Although the target of CRT is electromechanical dyssynchrony, the ability to identify both electrical dyssynchrony (ED) and mechanical dyssynchrony (MD) is challenging. ED can be defined using vectorcardiographic QRS area.⁴ Previously, we have demonstrated that ED (defined by QRS area ≥ 95 μVS)⁴ is independent from MD (defined as the classical LBBB contraction pattern)5, 6, 7 and the co-presence of ED and MD is linked to a more favorable CRT outcome⁸; however, factors that may influence the presence or absence of ED and MD, particularly in patients with borderline QRSd, remain poorly understood. In this study, we aim to investigate the associations between clinical characteristics, such as sex, height, and underlying cardiomyopathy substrates, and ED and MD. We hypothesized that among patients with LBBB and borderline QRSd, QRS area indexed to left ventricular end-diastolic volume (QRSa/LVEDV) would better predict dyssynchrony and clinical outcomes than QRSd alone, potentially offering a more individualized approach to CRT patients.

Methods

We analyzed data in this single-center retrospective cohort study, consisting of a total of 336 patients with a CRT device from April 2006 to September 2015 with LBBB and evaluable digital electrocardiograms (ECGs) and transthoracic echocardiograms. The study adhered to the Helsinki Declaration and was approved by the Duke Institutional Review Board, which waived the need for informed consent.

ECG analyses

Clinically obtained digital standard 12-lead ECGs were reanalyzed in the GE MUSE Cardiology Information System version 8.0.2.10132 with analysis software version 241 (GE Healthcare, Chicago, IL) and exported in XML format. 2 investigators blinded to outcomes (DJF and KE) manually overread the fiducial points used to measure QRSd and adjudicated QRS morphology, including differentiation between strict and non-strict LBBB using the Strauss criteria.⁹

Vectorcardiograms were derived from the XML files using customized MATLAB software (MathWorks, Inc., Natick, MA) using the Kors matrix.¹⁰ We calculated the QRSa^4,11 using the median complex. The integral for the depolarization curves representing the X, Y, and Z planes and the 3-dimensional QRSa was calculated as (QRSx² + QRSy² +QRSz²)^1/2, where QRSx, QRSy, and QRSz were the integrals of the depolarization curves in the X, Y, and Z planes (Figure 1A).¹¹ ED was defined as a QRSa of ≥95 μVs, based on prior work demonstrating this threshold was able to discriminate long-term outcomes among CRT patients.⁴

Figure 1.

A: QRS area calculation. B: Classical pattern assessment. QRS area is calculated by measuring the absolute value of the area under the curve in the X, Y, and Z planes from the vectorcardiogram. These individual values are squared and summed; the QRS area is finally calculated by taking the square root of this sum.⁴ The classical left bundle branch block contraction pattern is assessed by comparing the contraction pattern of the septal and lateral walls using segmental longitudinal strain from an apical 4-chamber view. With the classical pattern, the lateral wall (blue line) stretches early (1) and contracts late (2) after aortic valve closure; the septal wall (green line) contracts early (3) and peaks within the first 70% of ejection.¹² (Figures reprinted with permission)

Echocardiography

Clinical echocardiograms were re-analyzed by a blinded investigator (KE) and verified by a second investigator (PS). Analyses were performed in ImageArena version 4.6 (TomTec Imaging Systems, Unterschleissheim, Germany) as previously described (Figure 1B).^5,12 LV volumes were calculated based on a modified Simpson’s triplane method. Speckle-tracked longitudinal strain patterns were generated and classified as either a typical LBBB contraction pattern (MD) or “other” contraction pattern (not considered dyssynchrony). The clinical implementation pathway was proposed in Figure 2.^6,7

Figure 2.

Clinical implementation pathway. Schematic of the QRS area/left ventricular end-diastolic volume (QRSa/LVEDV) computation pipeline: standard 12-lead electrocardiograms (ECGs) were exported in XML format from GE MUSE, transformed into vectorcardiograms using the Kors regression matrix as previously published in the open-source BRAVEHEART software package,²⁶ and the resulting QRSa was divided by echocardiographic LVEDV to yield the QRSa/LVEDV ratio, with the prespecified optimal threshold of 0.375 applied (created by biorender.com). ECG = electrocardiogram; LVEDV = left ventricular end-diastolic volume; QRSa = QRS area.

Outcomes

The study endpoint was a composite of HF hospitalization, left ventricular assist device (LVAD) surgery, cardiac transplantation, or all-cause death. Patients who did not experience this endpoint by the query date were censored at the query date. Secondary endpoints included all-cause mortality alone and HF hospitalization alone.

Statistical analysis

Continuous variables are presented as mean ± standard deviation, while categorical variables are represented as count (%). Differences between groups were evaluated using the Student t-test for continuous variables, and the chi-square (χ²) test for categorical variables.

Logistic regression models were used to evaluate the association between clinical and demographic variables and dyssynchrony (MD and ED). Univariable models were first constructed for each predictor. Multivariable models were then developed, including variables with a univariable P-value of <.20, using stepwise selection with a significance level of P < .05 to remain in the final model. Covariates included age, sex, height, body mass index, QRSd, LVEF at baseline, history of ischemic heart disease, atrial fibrillation, and New York Heart Association (NYHA) functional class as previously described.¹³ To evaluate whether height’s association with dyssynchrony was mediated by heart size, we performed sensitivity analyses substituting LVEDV for height in multivariable models. Odds ratios (ORs), 95% confidence intervals (CIs), and P-values were reported for each model.

Given the correlation between sex and height, collinearity was assessed using variance inflation factors (VIF), with VIF of <5 considered acceptable. Sex and height interactions were tested for ED, MD, and clinical outcomes.

Model discrimination was evaluated using the area under the receiver operating characteristic (ROC) curve and Somers’ D statistic, with the optimized threshold value defined by the maximum Youden’s index to balance sensitivity and specificity. Separate models were also constructed for subgroups stratified by height tertiles and QRSd based on the 150 ms threshold to assess potential effect modification.

For time-to-event analyses, the primary composite outcome was defined as either HF hospitalization, LVAD implantation, cardiac transplantation, or all-cause death. Cox proportional hazards models were used to estimate hazard ratios (HRs) and 95% CIs for each variable, with separate models constructed for patients with QRSd of <150 ms and ≥150 ms. For a sensitivity analysis examining HF hospitalization as a separate endpoint, Fine–Gray competing risk regression was performed with death treated as a competing event. Kaplan–Meier curves were generated to visualize event-free survival stratified by QRSd within tertiles of QRSa/LVEDV, and differences between groups were assessed using log-rank and Wilcoxon tests.

The primary hypothesis that QRSa/LVEDV would show better discrimination was specified a priori, as was the composite endpoint of HF hospitalization, LVAD implantation, cardiac transplantation, or all-cause death. In contrast, the comparison with QRSd/LVEDV was post hoc due to recently published data.

Results

Study cohort

A total of 336 patients with LBBB were included in this retrospective analysis of the Duke CRT database from April 2006 to September 2015 (Table 1). The mean age was 65.8 ± 11.4 years, and 204 (60.7%) of the cohort were men. Other baseline characteristics are presented in Table 1. Notably, hypertension (70.2%), non-ischemic dilated cardiomyopathy (58.0%), and ischemic heart disease (47.9%) were the most common comorbidities. Most patients experienced NYHA class III symptoms (75.3%) and had a HF duration of over 9 months (80.1%). The use of guideline-directed medical therapy was common, including beta-blockers (90%), diuretics (83%), and angiotensin-converting enzyme inhibitors (61%).

Table 1.

General characteristics

Characteristic	Registry patients (N = 336)	QRS < 150 (N=139)	QRS ≥ 150 (197)	P-value
Demographics
Mean age: years ± SD	65.8 ± 11.4	65.10 ± 11.47	66.23 ± 11.38	.372
Sex: number (%)
Male	204 (60.7)	87 (62.6)	117 (59.4)	.633
Race and ethnicity: number (%)
American Indian or Alaska Native	4 (1.2)	1 (0.7)	3 (1.5)	.445
Asian	2 (0.6)	1 (0.7)	1 (0.5)
Black or African American	73 (21.7)	31 (22.3)	42 (21.3)
White	168 (50.0)	76 (54.7)	92 (46.7)
Unknown	86 (25.6)	28 (20.1)	58 (29.4)
Other race	3 (0.9)	2 (1.4)	1 (0.5)
Medications at enrollment: number (%)
Angiotensin-converting enzyme inhibitor (ACEI)	206 (61%)	82 (59.0)	124 (62.9)	.536
Angiotensin receptor blocker (ARB)	67 (20%)	27 (19.4)	40 (20.3)	.952
Amiodarone	53 (16%)	21 (15.1)	32 (16.2)	.897
Beta Blocker	301 (90%)	125 (89.9)	176 (89.3)	1.000
Coumadin	74 (22%)	39 (28.1)	35 (17.8)	.035
Digoxin	55 (16%)	29 (20.9)	26 (13.2)	.085
Diuretic	280 (83%)	112 (80.6)	168 (85.3)	.322
Mexiletine	1 (0.3%)	1 (0.7)	0 (0.0)	.861
Sotalol	6 (2%)	2 (1.4)	4 (2.0)	1.000
Other Clinical Diagnosis
Atrial Fibrillation or Flutter	93 (27.7%)	45 (32.4)	48 (24.4)	.136
Cardiac Arrest	22 (7%)	10 (7.2)	12 (6.1)	.858
Cerebrovascular Disease	39 (11.6%)	16 (11.5)	23 (11.7)	1.000
Chronic Lung Disease	62 (18.5%)	32 (23.0)	30 (15.2)	.095
Diabetes	101 (30.0%)	43 (30.9)	58 (29.4)	.862
Hypertension	236 (70.2%)	102 (73.4)	134 (68.0)	.349
Ischemic Heart Disease	161 (47.9%)	73 (52.5)	88 (44.7)	.191
Myocardial Infarction	121 (36.0%)	53 (38.1)	68 (34.5)	.573
Non-ischemic Dilated Cardiomyopathy	195 (58.0%)	68 (48.9)	127 (64.5)	.006
Syncope	35 (10%)	12 (8.6)	23 (11.7)	.473
Ventricular Tachycardia	56 (16.7%)	30 (21.6)	26 (13.2)	.060
Prior Intervention
Coronary Artery Bypass Graft	87 (25.9%)	40 (28.8)	47 (23.9)	.375
Dialysis	12 (3.6%)	6 (4.3)	6 (3.0)	.749
Implantable Cardioverter-Defibrillator	55 (16.4%)	30 (21.6)	25 (12.7)	.043
Percutaneous Coronary Intervention	79 (24%)	31 (22.3)	48 (24.4)	.758
Pacemaker	17 (5.1%)	9 (6.5)	8 (4.1)	.458
NYHA Class
Class I	12 (3.6%)	5 (3.6)	7 (3.6)	.629
Class II	57 (17.0%)	21 (15.1)	36 (18.3)
Class III	253 (75.3%)	109 (78.4)	144 (73.1)
Class IV	14 (4.2%)	4 (2.9)	10 (5.1)
QRS Morphology
Strict LBBB	265 (78.9%)	71 (51.1)	194 (98.5)	<.001
Non-strict LBBB	71 (21.1%)	68 (48.9)	3 (1.5)
Symptoms of Heart Failure Onset
<3 months	18 (5.4%)	7 (5.2)	11 (5.6)	.252
3–9 months	44 (13.1%)	13 (9.6)	31 (15.8)
>9 months	269 (80.1%)	115 (85.2)	154 (78.6)
Dyssynchrony
Electrical Dyssynchrony (QRSa>=95 μVS)	178 (53.0%)	34 (24.5)	144 (73.1)	<.001
Classical Mechanical Dyssynchrony	250 (74.4%)	86 (61.9)	164 (83.2)	<.001

ACEI = angiotensin-converting enzyme inhibitor; ARB = angiotensin receptor blocker; LBBB = left bundle branch block; NYHA = New York Heart Association; QRSa = QRS area.

Median follow-up was 28.3 months (interquartile range: 9.0–68.4). During follow-up, 241 patients (71.7%) experienced a composite endpoint, including 149 exacerbated HF hospitalizations, 205 deaths, and 46 LVAD implantations or heart transplantations. Event rate was similar between QRS subgroups: 74.8% (QRS <150 ms) and 69.5% (QRS ≥150 ms).

ED, defined by QRSa of ≥95 μVS, was present in 178 patients (53.0%), while MD, defined as the classical LBBB contraction pattern, was observed in 250 patients (74.4%). Both ED and MD were less prevalent among patients with QRS <150 ms compared to ≥150 ms (ED: 24.5% vs 73.1%; MD: 61.9% vs 83.2%).

Predictors of electrical dyssynchrony

ED was less common among patients with a QRSd of <150 ms compared to ≥150 ms (24.5% vs 73.1%, P < .0001). Among patients with a QRSd of <150 ms, increasing height was associated with lower odds of ED (OR: 0.928; 95% CI: 0.891–0.966; P = .0003), and female sex was associated with higher odds (OR: 3.900; 95% CI: 1.815–8.382; P = .0005), yet only height was statistically significant in the multivariable analysis (OR: 0.928; 95% CI: 0.891–0.966; P = .0003; Table 2). Among patients with QRSd of ≥150 ms, none of the variables examined were associated with ED in either univariable or multivariable models.

Table 2.

Predictors of electrical dyssynchrony

QRS < 150
Variable	Univariable Model			Multivariable Model
	OR	CI	P-value	OR	CI	P-value
Age	1.003	0.971–1.037	.8403			NS
Afib flutter	0.914	0.415–2.009	.8221			NS
Ischemic HD	0.373	0.173–0.802	.0116			NS
Baseline LVEF	5.366	0.106–272.11	.4016			NS
Diabetes	0.842	0.377–1.882	.6749			NS
Height	0.928	0.891–0.966	.0003	0.928	0.891–0.966	.0003
NYHA	0.778	0.405–1.495	.4505			NS
QRS Morphology	0.429	0.197–0.931	.0323			NS
Female Sex	3.900	1.815–8.382	.0005			NS
LVEDV	0.995	0.989–1.001	.1025			NS
QRS ≥ 150
	Univariable Model			Multivariable Model
Variable	OR	CI	P-value	OR	CI	P-value
Age	1.002	0.974–1.032	.8648			NS
Afib flutter	0.697	0.333–1.460	.3388			NS
Ischemic HD	0.782	0.405–1.509	.4637			NS
Baseline LVEF	1.864	0.043–80.632	.7460			NS
Diabetes	0.533	0.267–1.065	.0750			NS
Height	1.015	0.981–1.051	.3867			NS
NYHA	0.729	0.398–1.338	.3081			NS
QRS Morphology	0.167	0.015–1.881	.1474			NS
Female Sex	0.945	0.485–1.839	.8668			NS
LVEDV	1.003	0.999–1.008	.1027			NS

ARB = angiotensin receptor blocker; CI = confidence interval; HD = heart disease; LVEF = left ventricular ejection fraction; LVEDV = left ventricular end-diastolic volume; NYHA = New York Heart Association; OR = odds ratio.

Predictors of mechanical dyssynchrony

MD was less common among patients with a QRSd of <150 ms compared to ≥150 ms (61.9% vs 83.2%, P < .0001). Among these with a QRSd of <150 ms, several variables were associated with MD in the univariable models (height, sex, ischemic heart disease, and strict versus non-strict LBBB); only female sex was independently associated (OR: 3.153; 95% CI: 1.479–6.725; P = .0030; Table 3) in the multivariable model. Among these with QRSd of ≥150 ms, age, ischemic heart disease and diabetes were significant in univariable analyses, and remained significant in the multivariable analysis (age: OR: 0.955; 95% CI: 0.913–0.999; P = .0454; ischemic heart disease: OR: 0.398; 95% CI: 0.169–0.938; P = .0352; diabetes: OR: 0.200; 95% CI: 0.086–0.467; P = .0002), while height, sex, and strict versus non-strict LBBB were not significant.

Table 3.

Predictors of mechanical dyssynchrony with and without QRSa/LVEDV

QRS < 150
Variable	Univariable Model			Multivariable Model (with QRSa/LVEDV)			Multivariable Model (without QRSa/LVEDV)
	OR	CI	P-value	OR	CI	P-value	OR	CI	P-value
QRSa/LVEDV	26.83	4.398–163.67	.0004	26.83	4.398–163.67	.0004	NA	NA	NA
Age	0.996	0.967–1.026	.8141			NS			NS
Afib flutter	0.574	0.284–1.159	.1215			NS			NS
Ischemic HD	0.499	0.253–0.984	.0448			NS			NS
Baseline LVEF	5.602	0.140–224.02	.3599			NS			NS
Diabetes	0.514	0.253–1.047	.0669			NS			NS
Height	0.954	0.922–0.987	.0068			NS			NS
NYHA	0.851	0.454–1.597	.6162			NS			NS
QRS Morphology	0.451	0.229–0.888	.0213			NS			NS
Female Sex	3.153	1.479–6.725	.0030			NS	3.153	1.479–6.725	.0030
LVEDV	0.993	0.988–0.998	.0093			NS			NS
QRS ≥ 150
	Univariable Model			Multivariable Model (with QRSa/LVEDV)			Multivariable Model (without QRSa/LVEDV)
Variable	OR	CI	P-value	OR	CI	P-value	OR	CI	P-value
QRSa/LVEDV	0.596	0.155–2.289	.4512			NS	NA	NA	NA
Age	0.957	0.920–0.994	.0245	0.955	0.913–0.999	.0454	0.955	0.913–0.999	.0454
Afib flutter	0.894	0.369–2.166	.8038			NS			NS
Ischemic HD	0.393	0.176–0.876	.0224	0.398	0.169–0.938	.0352	0.398	0.169–0.938	.0352
Baseline LVEF	0.666	0.008–52.457	.8552			NS			NS
Diabetes	0.225	0.101–0.501	.0003	0.200	0.086–0.467	.0002	0.200	0.086–0.467	.0002
Height	0.991	0.951–1.032	.6603			NS			NS
NYHA	0.802	0.397–1.623	.5402			NS			NS
QRS Morphology	1.438	0.046–45.027	.8363			NS			NS
Female Sex	2.222	0.936–5.274	.0703			NS			NS
LVEDV	1.005	1.000–1.010	.0622			NS			NS

ARB = angiotensin receptor blocker; CI = confidence interval; HD = heart disease; LVEF = left ventricular ejection fraction; LVEDV = left ventricular end-diastolic volume; NYHA = New York Heart Association; OR = odds ratio; QRSa = QRS area.

Clinical outcomes

Due to the complex relationship between body size, sex, and dyssynchrony measures, we sought to identify a single variable that would best predict outcomes after CRT, particularly among patients with a QRSd of <150 ms. Sequential models relating indexed QRS variables and the risk of the composite clinical endpoint are summarized in Table 4. In the final model, QRSa/LVEDV was the strongest predictor of risk of HF hospitalization, LVAD, transplant, or death in both QRS of <150 ms (HR: 0.172; 95% CI: 0.063–0.470; P = .0006) and QRS of ≥150 ms subgroups (HR: 0.407; 95% CI: 0.187–0.886; P = .0235), while QRSd alone was not predictive. Importantly, QRSa/LVEDV was the only QRS variable that rendered sex a non-significant predictor among those with a QRSd of <150 ms. Therefore, QRSa/LVEDV is strongly and inversely related to the risk of the composite endpoint among patients with LBBB and a QRSd of <150 ms. Kaplan–Meier analysis confirmed that QRSa/LVEDV tertiles were associated with event-free survival in both QRSd subgroups (QRS < 150 ms: P = .0087; QRS ≥ 150 ms: P = .0001) (Figure 3).

Table 4.

Independent association between QRS variables, Sex, and risk of HF hospitalization, LVAD, cardiac transplantation, or death

QRSd <150
	QRS Variable			Sex
	HR	CI	P-value	HR	CI	P-value
QRSd			NS	0.596	0.391–0.911	.0167
QRSd/Height			NS	0.554	0.342–0.896	.0160
QRSd/LVEDV			NS		NS
QRSd/BSA			NS	0.601	0.376–0.960	.0329
QRSa	0.991	0.983–0.999	0.0322			NS
QRSa/Height	0.236	0.062–0.896	0.0339			NS
QRSa/LVEDV	0.255	0.080–0.814	0.0210			NS
QRSa/BSA			NS			NS
QRSd ≥150
	QRS Variable			Sex
	HR	CI	P-value	HR	CI	P-value
QRSd			NS			NS
QRSd/Height			NS			NS
QRSd/LVEDV			NS			NS
QRSd/BSA			NS			NS
QRSa	0.994	0.990–0.999	.0176			NS
QRSa/Height	0.362	0.159–0.824	.0155			NS
QRSa/LVEDV	0.399	0.188–0.850	.0172			NS
QRSa/BSA			NS			NS

BSA = body surface area; CI = confidence interval; HR = hazard ratio; LVEDV = left ventricular end-diastolic volume; QRSa = QRS area; QRSd = QRS duration.

Figure 3.

Kaplan–Meier curves by QRS duration groups (<150 ms and ≥150 ms). Survival stratified by QRSa/LVEDV tertiles within patients with QRS duration of <150 ms and ≥150 ms. LVEDV = left ventricular end-diastolic volume; QRSa = QRS area.

Among patients with QRSd of <150 ms, ROC analysis demonstrated that QRSa/LVEDV had significantly better discrimination for MD than QRSd/LVEDV^14,15 (AUC: 0.745 vs 0.654; DeLong P = .019; Figure 4). QRSa/LVEDV and QRSd/LVEDV were moderately correlated (r = 0.68; P < .001). Similarly, QRSa/LVEDV was superior to QRSd/LVEDV for the prediction of the composite outcome (c-index: 0.574 vs 0.554). In a model including both variables, only QRSa/LVEDV remained significant (HR: 0.146; 95% CI: 0.037–0.581; P = .006), while QRSd/LVEDV was not (HR: 2.17; 95% CI: 0.73–6.41; P = .16), confirming that QRS area provides incremental prognostic information beyond QRSd when indexed to ventricular size.

Figure 4.

QRSa/LVEDV receiver operating characteristic (ROC) curve for MD: Receiver operating characteristic curves evaluate model performance for mechanical dyssynchrony. The final multivariable model (blue line) achieved an area under the curve of 0.745. AUC= area under the curve; LVEDV = left ventricular end-diastolic volume; QRSa = QRS area; ROC = receiver operating characteristic.

We examined whether the prognostic value of QRSa/LVEDV is mediated through its association with MD. MD alone was a significant predictor of the composite endpoint (HR: 0.51; 95% CI: 0.33–0.78; P = .002). When MD was added to the model containing QRSa/LVEDV and sex, MD remained independently associated with outcomes (HR: 0.60; 95% CI: 0.38–0.93; P = .024), while the QRSa/LVEDV association was attenuated and no longer reached statistical significance (HR: 0.35; 95% CI: 0.11–1.09; P = .070). The correlation between MD and QRSa/LVEDV was modest (r = 0.35 in QRS < 150 ms; r = −0.07 in QRS ≥ 150 ms), while VIF was low. Kaplan–Meier analyses demonstrated that within tertiles of QRSa/LVEDV, QRSd was not significantly associated with clinical outcomes in any tertile (all P > .05) (Supplemental Figure 1).

Finally, we examined whether the composite endpoint masked differential effects on its individual components. Among patients with QRSd of <150 ms, there were 90 deaths and 64 HF hospitalizations during follow-up. Additionally, higher QRSa/LVEDV was associated with reduced risk of both endpoints. For HF hospitalization, the association was statistically significant (HR: 0.13; 95% CI: 0.03–0.60; P = .009); this finding remained consistent in the Fine–Gray model accounting for death as a competing event (subdistribution HR: 0.14; 95% CI: 0.04–0.58; P = .007). For all-cause mortality, the effect was in the same direction but did not reach statistical significance (HR: 0.40; 95% CI: 0.14–1.19; P = .10; Supplemental Table 1).

Discussion

In this retrospective analysis of patients undergoing CRT, both sex and height were associated with measures of ED and MD, although their relative importance differed depending on the type of dyssynchrony. Importantly, height emerged as an independent predictor of ED, while female sex was more strongly associated with MD. Additionally, among patients with a QRSd of <150 ms, QRSa/LVEDV was a robust predictor of clinical adverse outcomes, outperforming QRSd, QRSd/LVEDV, sex, and height.

Prior studies have shown that women tend to derive greater benefit from CRT than men.¹⁶ Our results demonstrating that women with QRSd of <150 ms are more likely to exhibit MD are consistent with prior observations that women might respond more favorably to CRT.^17,18 More recent patient-level meta-analyses suggest that CRT may also be beneficial in individuals with shorter stature, even when QRSd falls below 150 ms. This has led to proposals to index QRSd to body or heart size as a potentially more accurate method for identifying appropriate CRT candidates among patients with borderline QRSd.¹³ Surrogates for heart size, including sex and height, have previously been used to explore the relationship between body size and heart size, but adverse remodeling in HF could alter heart size independent of sex and height. The present study’s findings suggest that patient height contributes meaningfully to ED when QRSd is modest but not at longer QRSd. In contrast, once ED is normalized for ventricular size, the predictive influence of sex is substantially attenuated, which underscores the hypothesis that myocardial mass is at least partially involved in the sex-based differences in CRT outcomes. While Maass et al¹⁹ demonstrated that QRS area alone was sufficient to remove sex as an independent predictor in multivariate analysis, our findings suggest that further indexing QRSa to LVEDV can provide additional discriminatory value for predicting MD and clinical outcomes in patients with QRSd of <150 ms. This finding also highlights that QRSa/LVEDV may provide a more equitable and individualized approach to CRT candidacy assessment. Of note, in Salden et al,²⁰ both female sex and QRSa/LVEDV were significant predictors in patients of all QRS morphologies and durations, with endpoints being LVAD, transplant, or death. In contrast, our study focused specifically on LBBB patients and stratified QRSd to align with guideline-based clinical decision-making. We also included HF hospitalization in our composite endpoint.²⁰

Further complicating CRT selection is the interaction between QRS morphology and underlying cardiomyopathy substrate, including more favorable outcomes from patients with non-ischemic cardiomyopathy²¹ and strict LBBB morphology.²² Building upon prior work, our results suggest that indexing QRSa to myocardial size better characterizes the substrate for CRT benefit. Yamamoto et al¹⁴ recently demonstrated that QRSd indexed to LVEDV predicted outcomes in patients with mid-range QRSd. Our findings extend this work by showing that QRSa/LVEDV provides improved discrimination compared to QRSd/LVEDV for predicting mechanical dyssynchrony, likely because QRS area captures 3-dimensional electrical information that QRSd alone cannot. In patients with a QRSd of <150 ms, in whom CRT decisions remain debated, QRSa/LVEDV was associated with differential risk for adverse clinical events, including HF hospitalization, LVAD implantation, heart transplant, or death. While these findings suggest that indexing electrical activation to myocardial size may improve substrate characterization, they do not establish the degree of benefit derived from CRT, as their outcomes might be even worse without therapy.

We also explored whether the prognostic value of QRSa/LVEDV operates through its association with MD. When MD was added to models containing QRSa/LVEDV, the HR for QRSa/LVEDV was attenuated by 16% and no longer reached statistical significance, while MD remained independently predictive. This suggests that QRSa/LVEDV predicts outcomes largely through its ability to identify patients with mechanical dyssynchrony, confirming a possible biological mechanism, yet both variables capture complementary information, as QRSa/LVEDV likely reflects electrical substrate while MD reflects mechanical consequences; however, the broad clinical application of echocardiographic MD assessment remains challenging due to technical difficulties and reproducibility concerns.²³

Within patient subgroups stratified by a 150 ms QRSd threshold, QRSd was not an independent predictor of clinical outcomes. By contrast, QRSa/LVEDV retained prognostic utility in both short and long QRSd subgroups, suggesting that global measures of electrical activation scaled to myocardial size are superior to QRSd in isolation. Specifically, it accounts for both the degree of electrical delay and the size of the substrate. For instance, a large QRSa coupled with a small LVEDV could be a reflection of true ED in a ventricle that still has mechanical reserve, a case more likely to benefit from CRT. In contrast, patients with a large LVEDV may have prolonged activation time simply due to increased size rather than pathological dyssynchrony, which means they may be less responsive to CRT despite a high QRSa. Similarly, a low QRSa regardless of LVEDV may indicate minimal dyssynchrony, and thus a possible limited role for resynchronization. Taken together, QRSa/LVEDV may help distinguish patients whose QRS prolongation reflects true electrical-mechanical mismatch from those in whom it may be a passive marker of chamber dilation. This finding is clinically relevant given ongoing debates about CRT use in patients with intermediate QRSd, where current guidelines remain cautious and individualized assessment is encouraged. In an era that increasingly focuses on precision HF therapies, metrics such as QRSa/LVEDV may offer a path toward refining CRT utilization.

Additionally, the study period preceded contemporary guideline-directed medical therapy, including sacubitril/valsartan and SGLT2 inhibitors; the interaction between these therapies and QRSa/LVEDV predictive value warrants investigation. Specifically, pivotal trials of sacubitril/valsartan and SGLT2 inhibitors reported absolute reductions in the combined endpoint of cardiovascular death or HF hospitalization of approximately 4–5 percentage points each, suggesting the absolute event rate in a fully optimized contemporary cohort would likely be considerably lower. We expect that the relative discrimination of QRSa/LVEDV would be preserved given that it reflects underlying electromechanical substrate rather than treatment response, but the absolute risk separation across tertiles would likely be attenuated.

Limitations

Our study has several limitations that should be noted. First, the retrospective design and single-center setting may limit the generalizability of our findings. Second, although our cohort is relatively large for a detailed CRT analysis, subgroup analyses may still be underpowered, as evidenced by the consistent direction of effect but lack of statistical significance for mortality alone in the QRS of <150 ms subgroup. Third, patients were programmed with biventricular pacing until the adaptive CRT algorithm became available, at which point it was typically programmed on. Quadripolar leads and fusion pacing algorithms were also consistently used once clinically available. At Duke, CRT optimization was not routinely performed with echocardiography, consistent with current clinical standards. Although device-based algorithms such as SMART-AV²⁴ and AdaptivCRT²⁵ held promise, clinical trials of these algorithms have been overall disappointing. Since the study period, conduction system pacing (His bundle pacing, left bundle branch area pacing) has emerged as an alternative, and whether QRSa/LVEDV retains predictive value for these modalities requires further study. Finally, while QRSa/LVEDV demonstrated a strong predictive performance, prospective validation in larger and multicenter cohorts is necessary. In addition, since no patients without CRT were examined here, the impact of CRT itself on any of these subpopulations is speculative.

Conclusion

QRSa/LVEDV emerged as a strong, individualized predictor of adverse outcomes, particularly among patients with QRSd of <150 ms. Other markers, such as QRSd alone and patient sex, were less predictive once heart size was accounted for. Our results suggest that indexing QRSa to ventricular size may improve CRT selection and risk stratification, especially in borderline patients, but without a non-CRT comparator arm, QRSa/LVEDV can be established only as a prognostic marker and not as a predictor of CRT-specific benefit. Prospective studies incorporating non-CRT controls are needed to guide CRT implantation decisions more effectively.

Disclosures

Dr Friedman has received: research grants from Abbott, Boston Scientific, Medtronic, and Philips, and consulting fees from Abbott, Alleviant, Boston Scientific, and Medtronic. Dr Loring receives grant support from the Department of Veterans Affairs and serves as a consultant for Huxley Medical Inc, and Boston Scientific.