Left Ventricular Mechanical Dispersion and Its Value in Combination With Global Longitudinal Strain to Predict Reverse Remodeling in Patients With Heart Failure With Reduced Ejection Fraction

Abstract

Background

Left ventricular mechanical dispersion (LV‐MD), an indicator of LV dyssynchrony, is not well studied in heart failure with reduced ejection fraction (HFrEF). This study aims to investigate the profile of LV‐MD in HFrEF, and evaluate whether it provides predictive information in terms of LV reverse remodeling (LVRR), particularly, on top of LV global longitudinal strain (LV‐GLS).

Methods

In this observational cohort study, 239 consecutive patients hospitalized for HFrEF from 2020 to 2022 were included, and baseline and follow‐up echocardiographic studies were performed. By strain analysis, LV‐MD was calculated as the SD of time to peak negative longitudinal strain from 17 LV segments. LVRR was defined as an increase in LVEF of ≥10% and a decrease in LV end‐diastolic diameter index of ≥10% or ≤33 mm/m².

Results

During a median follow‐up of 19 months, LVRR occurred in 152 patients (64%). A great heterogeneity of LV‐MD ranging from 30 to 202 milliseconds was present in the cohort and 94 (84%) in 112 patients who were with LV‐MD >76 milliseconds had no left bundle‐branch block. LV‐MD ≤76 milliseconds independently predicted LVRR and possessed greater discriminatory ability than LV‐GLS by receiver operating characteristics analysis. Moreover, addition of LV‐MD ≤76 milliseconds and LV‐GLS ≤−5.5% to a multivariable model, including age, HF duration, systolic blood pressure, QRS duration and use of angiotensin receptor‐neprilysin inhibitors, improved predictive performance in terms of LVRR.

Conclusions

LV mechanical dyssynchrony is widely present in HFrEF, irrespective of left bundle‐branch block. LV‐MD is significantly associated with LVRR, independent of and superior to LV‐GLS. Combined assessments of LV‐MD and LV‐GLS provide additive value beyond clinical variables for predicting LVRR in HFrEF.

Nonstandard Abbreviations and Acronyms

GLS

global longitudinal strain

LVRR

left ventricular reverse remodeling

MD

mechanical dispersion

Clinical Perspective.

What Is New?

• In our cohort of patients with heart failure with reduced ejection fraction, although a great heterogeneity of mechanical dyssynchrony as assessed by left ventricular mechanical dispersion (LV‐MD) was observed, a great proportion of patients (47%) had prolonged LV‐MD (>76 milliseconds), of whom 84% had no left bundle‐branch block.

• Prolonged LV‐MD (>76 milliseconds) was independently associated with lower likelihood of LV reverse remodeling, and LV‐MD was superior to LV global longitudinal strain (LV‐GLS) for predicting LV reverse remodeling in heart failure with reduced ejection fraction; a multivariable model including clinical parameters, LV global longitudinal strain, and LV‐MD was built with improved predictive performance in terms of LV reverse remodeling.

What Are the Clinical Implications?

• In addition to traditional clinical predictors, adding LV‐MD and LV global longitudinal strain measured by speckle‐tracking echocardiography can improve LV reverse remodeling prediction and thus facilitate risk stratification and aid clinical decision making in heart failure with reduced ejection fraction; patients who would benefit from medical therapy may be identified, among whom more aggressive treatment as device implantation and heart transplantation may be delayed or avoided.

Heart failure (HF) is a major public health concern with substantial morbidity and mortality. ¹ Cardiac remodeling, manifested as adverse changes of cardiac chamber size, shape and function, is the hallmark of HF development and progression and associated with poor outcomes. By far it is best studied in HF with reduced ejection fraction (HFrEF). ² On the other hand, such changes may be reverted toward that of a normal heart, termed as left ventricular reverse remodeling (LVRR), indicating better prognosis. ³ , ⁴

Currently, we still lack enough understanding about the LVRR process. In studies of HFrEF, the incidence of LVRR varies from 30% to 50%. ³ , ⁴ , ⁵ , ⁶ In patients with LVRR, the magnitude of improvement is greatly inconsistent across patients, and the timing of LVRR also remains unclear, from as early as 3 months to up to 2 years after initiation of treatment. ⁷ Therefore, early prediction of LVRR is of pivotal importance for both guiding more tailored therapy and outcome prediction. In patients with higher likelihood of LVRR, more aggressive therapeutic options, such as cardiac resynchronization therapy (CRT), LV assist device, and heart transplantation, may be avoided. Across numerous HF drug and device trials, a number of clinical and imaging parameters have been linked to LVRR. Predictors favoring LVRR include female sex, shorter HF duration, nonischemic cause, higher systolic blood pressure (SBP), lack of left bundle‐branch block (LBBB), lower concentrations of biomarkers, optimized medical therapy, and the absence or a lower extent of late gadolinium enhancement on cardiac magnetic resonance (CMR). ⁶ , ⁷ , ⁸ , ⁹

Echocardiography is the initial and main image tool in the management process of HF. However, the role of traditional parameters such as LV ejection fraction (LVEF) and LV chamber size in predicting LVRR remains controversial. ⁶ , ⁷ Novel imaging techniques for myocardial mechanics have expanded the prognostic role of echocardiographic parameters in HF. For example, speckle‐tracking echocardiography‐derived global longitudinal strain (GLS) has demonstrated superior prognostic value over LVEF. ¹⁰ , ¹¹ , ¹² Another strain‐derived parameter, LV mechanical dispersion (LV‐MD), is a quantitative measure of mechanical dyssynchrony, which has been linked to sudden cardiac death and ventricular arrhythmias in a variety of cardiovascular diseases including HF. ¹³ , ¹⁴ , ¹⁵ , ¹⁶ Although LV‐MD has not been well studied in HFrEF and its role in predicting LVRR is not fully clarified, preliminary studies have suggested its relevance to CRT response assessment. ¹⁷ Moreover, the utility of qualitative echocardiographic measures of mechanical dyssynchrony, such as septal flash and apical rocking, has been reported in LVRR prediction as well. ¹⁸ , ¹⁹ , ²⁰ Therefore, we hypothesize that LV‐MD is associated with both structural and functional LV remodeling and provides incremental value on predicting LVRR on top of LVEF and GLS.

METHODS

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Study Population

The present study was retrospective analysis of a single‐center prospective observational cohort. From January 2020 to December 2022, consecutive patients with HFrEF (LVEF ≤40%) who were admitted to cardiology department of Tongji Hospital were screened. The diagnosis of HFrEF was made according to the European Society of Cardiology guidelines. ²¹ Patients who were followed up for at least 12 months and attended the core echocardiography laboratory at the cardiology department were eligible for inclusion in the study. Exclusion criteria were as follows: (1) patient refusal; (2) patients with myocarditis, stress cardiomyopathy, hyperthyroidism, tachycardia‐induced cardiomyopathy, peripartum cardiomyopathy, and congenital and severe primary valvular heart diseases; and (3) insufficient echocardiographic image quality. Figure 1 demonstrated the patient selection and a total of 239 patients constituted the final study population.

Figure 1. Flow chart of the study population.

HFrEF indicates heart failure with reduced ejection fraction; and TTE, transthoracic echocardiography.

The baseline clinical and demographic characteristics were collected. The medical treatments were according to European Society of Cardiology HF guidelines ²¹ and at the discretion of each treating physician. The ischemic pathogenesis was defined by 1 of the following criteria: (1) significant epicardial coronary artery stenosis by coronary angiography (≥50% in the left main coronary artery, or ≥70% in the right, the left anterior descending and the left circumflex coronary artery); or (2) documented history of myocardial infarction or coronary revascularization.

This study was approved by the Institutional Review Board of Tongji Hospital in accordance with the Declaration of Helsinki. All participants provided written informed consent.

Echocardiography

At baseline and during follow‐up, all patients underwent a comprehensive transthoracic echocardiographic study on ultrasound systems (General Electric Vivid E95, GE Vingmed; Horten, Norway). Conventional echocardiographic measurements were performed according to established guidelines. ²² LV linear dimensions were measured from the parasternal long‐axis view. LV volumes and LVEF were measured from the apical 2‐ and 4‐chamber views, using the modified Simpson's biplane method. LV mass was estimated by the Devereux formula. The severity of functional mitral regurgitation was graded semiquantitatively as mild, moderate, or severe according to the jet area of color‐flow Doppler data.

Strain analysis was performed offline on a dedicated workstation (EchoPac version: 204; GE Vingmed; Horten, Norway) by an investigator blinded to the patient's clinical data. The endocardial border was automatically tracked, and the region of interest was adjusted to cover the entire myocardium. When regional tracking is suboptimal in >2 myocardial segments in a single view, the strain analysis was deemed inadequate and thus excluded from this study. LV‐GLS was defined as the average of peak systolic longitudinal strain values of 17 LV segments. LV‐MD was calculated as the SD of time from the onset of Q/R on the ECG to peak negative longitudinal strain from all 17 LV segments (Figure 2). To assess the interobserver reproducibility, stored echocardiograms from 20 randomly selected patients were studied by 2 independent readers. Intraobserver reproducibility was evaluated by asking each investigator to repeat strain analysis 1 week later, blinded to the previous measurements.

Figure 2. Longitudinal strain and mechanical dispersion in a patient with LVRR (A) and another without LVRR (B).

Two representatives of bull's‐eye display of GLS and longitudinal strain curves in the apical 4‐chamber view are shown. A, Strain output from a patient experiencing LVRR with LVEF 19%, LV‐GLS −5.5%, and LV‐MD 65 msec. B, Strain output from a patient experiencing no LVRR with LVEF 27%, LV‐GLS −7.9%, and LV‐MD 106 msec. Note the patient with LVRR had lower LV‐MD even with worse LVEF and GLS. Yellow vertical lines indicate onset of Q/R on the ECG. White horizontal arrows indicate contract duration (time to peak negative strain) in each LV segment. GLS indicates global longitudinal strain; LV, left ventricular; LVEF, left ventricular ejection fraction; LVRR, left ventricular reverse remodeling; and MD, mechanical dispersion.

Follow‐Up and Outcomes

The patients were followed up at our outpatient clinic at 1, 3, 6, 12, 18, and 24 months after discharge. The main outcome was LVRR determined by baseline and follow‐up echocardiography. LVRR was defined as the combined presence of (1) an absolute increase in LVEF of at least 10% and (2) a relative decrease in LV end‐diastolic diameter index of at least 10% or ≤33 mm/m², as previous reported. ⁴ , ⁵

Statistical Analysis

Continuous data are presented as mean±SD when normally distributed and as median and interquartile range (IQR) when not normally distributed. Categorical variables are expressed as absolute numbers and percentages. Continuous variables were compared between groups by Student t test or Mann–Whitney U test, whereas for binary variables, the Pearson chi square test or Fisher exact test was used. The Spearman rank correlation test was used to calculate the association between LV‐MD and other numeric variables in relation to LVRR. Receiver operating characteristic analysis was performed to assess the performance and determine the optimal threshold of variables for predicting LVRR. Continuous variables were made binary by the use of optimal cutoff values determined by receiver operating characteristic analysis. To identify the independent predictors of LVRR and adjust for confounding factors, variables with P<0.05 in univariable analysis and no significant collinearity were selected for inclusion in multivariable regression analysis, and the odds ratios (ORs) and 95% CIs were determined using logistic regression analysis. To inspect for multicollinearity, the Spearman rank correlation coefficient was calculated between baseline variables, assuming no significant multicollinearity when the correlation coefficient was <0.5. In addition, the variance inflation factor was also calculated, assuming no significant multicollinearity when this value was <5. To investigate the additive value of LV‐MD over clinical parameters and LV‐GLS to predict outcome, likelihood ratio test was performed, for which the global chi‐square values of models were calculated and reported. The C‐statistic was used to assess the discrimination of these prediction models. Calibration was assessed with the Hosmer–Lemeshow goodness‐of‐fit test. A 10‐fold cross‐validation was employed to assess the model's robustness. Moreover, sensitivity analysis was performed by incorporating patients lost to follow‐up, with missing data imputed using multiple imputation. Intra‐ and interobserver reproducibility of echocardiographic parameters were measured by calculating the intraclass correlation coefficients (ICCs). A 2‐tailed P value of <0.05 was considered for statistical significance. All statistical analyses were performed using SPSS software package (version 26.0, IBM SPSS Statistics, Armonk, NY, USA) and R version 4.4.1 (R Foundation for Statistical Computing, Vienna, Austria).

RESULTS

Baseline Characteristics

The baseline clinical, demographic, and echocardiographic characteristics of the study population were depicted in Table 1. Among the 239 patients in our cohort, the median age was 52 years (IQR, 39–63 years), 79% (n=190) were men, 34% (n=82) had an ischemic cause, and 8% (n=19) showed LBBB. The median LVEF was 26% (IQR, 20–30) and median LV end‐diastolic diameter index was 37 mm/m² (IQR, 33–41 mm/m²). The median LV‐GLS and LV‐MD were −6.5% (IQR, −8.2 to −5.0) and 73 milliseconds (IQR, 58–102 milliseconds), respectively. The distributions of LV‐GLS and LV‐MD were displayed in Figure S1, and a great heterogeneity of LV‐MD was present in the study cohort, ranging from 30 to 202 milliseconds. The Spearman rank correlation coefficient for LV‐MD and LV‐GLS was 0.353, whereas the variance inflation factor was 1.138, indicating no significant multicollinearity between the 2 variables.

Table 1.

Baseline Characteristics of the Overall Population

	Overall (n=239)
Age, y	52 (39–63)
Male sex	190 (79)
Body mass index, kg/m2	24.5 (22.5–27.7)
Heart rate, beats/min	88 (76–100)
Systolic BP, mm Hg	122 (107–138)
Diastolic BP, mm Hg	81 (71–92)
HF duration, mo	2.0 (0.5–12.0)
New York Heart Association functional class III/IV	172 (72)
Ischemic cause of HF	82 (34)
STEMI	12 (5)
Non‐STEMI	20 (8)
Other	50 (21)
Comorbidities
Hypertension	118 (49)
Diabetes	75 (31)
Chronic kidney disease	65 (27)
Dyslipidemia	101 (42)
ECG parameters
QRS duration, msec	104 (94–113)
Left bundle‐branch block	19 (8)
Atrial fibrillation/flutter	28 (12)
Nonsustained ventricular tachycardia	86 (36)
Serum markers
Troponin I, pg/mL	26 (12–102)
N‐terminal pro‐B‐type natriuretic peptide, pg/mL	2650 (1188–5371)
Echocardiographic parameters
LV ejection fraction, %	26 (20–30)
LV end‐diastolic diameter index, mm/m2	37 (33–41)
LV end‐systolic volume index, mL/m2	69 (53–88)
LV mass index, g/m2	158 (131–184)
Moderate/severe functional mitral regurgitation	59 (25)
LV‐GLS, %	−6.5 (−8.2 to −5.0)
LV‐GLS ≤−5.5%	157 (66)
LV‐MD, msec	73 (58–102)
LV‐MD ≤76 msec	127 (53)

Values are median (interquartile range) or n (%) for descriptive statistics. BP indicates blood pressure; GLS, global longitudinal strain; HF, heart failure; LV, left ventricular; MD, mechanical dispersion; and STEMI, ST‐segment–elevation myocardial infarction.

Among the study cohort, 217 (91%) were prescribed an angiotensin receptor‐neprilysin inhibitor (ARNI) or renin‐angiotensin system inhibitor, 209 (87%) a β‐blocker, 195 (82%) a mineralocorticoid receptor antagonist, and 85 (36%) an SGLT2i (sodium‐glucose cotransporter‐2 inhibitor). Besides, 72 patients underwent percutaneous coronary intervention. Table S1 showed the detailed information regarding treatments at baseline and during follow‐up. In contrast to patients lost to follow‐up, patients included in the analysis were younger, whereas most of clinical and echocardiographic variables, such as sex, pathogenesis, baseline LVEF, and HF duration, showed no statistical differences (Table S2). A total of 46 patients who died within 12 months were excluded from the study. Interestingly, their LV‐MD at baseline was significantly higher whereas LV‐GLS showed no difference, in comparison with the patients who completed 12‐month follow‐up and thus were included in the study (Table S3).

Outcomes and Predictors of LVRR

During a median follow‐up period of 19 months (IQR, 14–25 months), LVRR occurred in 152 patients (64%) at a median time of 7 months (IQR, 6–13 months). Significant differences between patients with and without LVRR were found with respect to age, body mass index, BP, HF duration, New York Heart Association functional class, ischemic cause, QRS duration, NT‐proBNP (N‐terminal pro‐B‐type natriuretic peptide), comorbidities, and drug use. The group with LVRR also exhibited higher baseline LVEF and LV‐GLS, lower LV dimensions/volumes and LV‐MD, and had lower proportion of moderate/severe functional mitral regurgitation (as shown in Table S4).

Baseline characteristics were evaluated for predicting LVRR by univariable and multivariable logistic regression analysis (Tables 2 and 3). Age≤55 years (OR, 2.87 [95% CI, 1.30–6.33], P=0.009), HF duration ≤2 months (OR, 2.48 [95% CI, 1.20–5.16], P=0.015), SBP ≥110 mm Hg (OR, 3.80 [95% CI, 1.70–8.51], P=0.001), QRS duration ≤110 milliseconds (OR, 2.40 [95% CI, 1.09–5.26], P=0.029), and ARNI use (OR, 2.77 [95% CI, 1.30–5.88], P=0.008) were found to be independent predictors of LVRR. For echocardiographic parameters, baseline LV‐GLS ≤−5.5% (OR, 2.12 [95% CI, 1.10–4.10], P=0.025) and LV‐MD ≤76 milliseconds (OR, 5.47 [95% CI, 2.91–10.29], P<0.001) were independently predictive of LVRR, and LVEF was excluded from the model partly due to the collinearity with LV‐GLS.

Table 2.

Clinical Predictors of the Achievement of LVRR by Univariable and Multivariable Logistic Regression Analysis

Baseline variable	Univariable		Multivariable*
	OR (95% CI)	P value	OR (95% CI)	P value
Age≤55 y	3.59 (2.07–6.24)	<0.001	2.87 (1.30–6.33)	0.009
Male sex	1.94 (1.02–3.66)	0.042
BMI ≥28 kg/m2	4.56 (2.04–10.18)	<0.001
Heart rate≥95 bpm	1.62 (0.91–2.87)	0.099
SBP ≥110 mm Hg	3.60 (2.00–6.48)	<0.001	3.80 (1.70–8.51)	0.001
DBP ≥80 mm Hg	3.01 (1.74–5.21)	<0.001
HF duration ≤2 mo	3.51 (2.02–6.09)	<0.001	2.48 (1.20–5.16)	0.015
NYHA functional class III/IV	0.40 (0.21–0.77)	0.006
Ischemic cause of HF	0.57 (0.33–0.99)	0.044
Hypertension	1.33 (0.79–2.26)	0.288
Diabetes	0.63 (0.36–1.09)	0.100
Chronic kidney disease	0.37 (0.20–0.66)	0.001
Dyslipidemia	1.67 (0.97–2.87)	0.067
QRS duration ≤110 msec	2.35 (1.32–4.18)	0.004	2.40 (1.09–5.26)	0.029
Left bundle‐branch block	0.48 (0.19–1.23)	0.125
Atrial fibrillation/flutter	0.53 (0.24–1.17)	0.116
NSVT	0.33 (0.19–0.59)	<0.001
Troponin I ≤75 pg/mL	2.20 (1.23–3.94)	0.008
NT‐proBNP ≤4500 pg/mL	3.04 (1.72–5.39)	<0.001
Angiotensin‐converting enzyme inhibitors/angiotensin receptor blockers	0.57 (0.31–1.04)	0.066
ARNI	2.80 (1.60–4.89)	<0.001	2.77 (1.30–5.88)	0.008
β‐blocker	2.22 (1.03–4.81)	0.043
Mineralocorticoid receptor antagonist	1.42 (0.73–2.76)	0.302
Diuretics	1.52 (0.71–3.27)	0.280
Sodium‐glucose cotransporter‐2 inhibitors	1.62 (0.92–2.85)	0.096

ARNI indicates angiotensin receptor‐neprilysin inhibitor; BMI, body mass index; DBP, diastolic blood pressure; HF, heart failure; LVRR, left ventricular reverse remodeling; NSVT, nonsustained ventricular tachycardia; NT‐proBNP, N‐terminal pro‐B‐type natriuretic peptide; NYHA, New York Heart Association; OR, odds ratio; and SBP, systolic blood pressure.

^*Variables included in multivariable logistic regression analysis were age, male, BMI, SBP, HF duration, NYHA functional class III/IV, ischemic cause of HF, chronic kidney disease, QRS duration, NSVT, troponin I, NT‐proBNP, ARNI, and β‐blocker. DBP was excluded from the model due to the collinearity with SBP.

Table 3.

Echocardiographic Predictors of the Achievement of LVRR by Univariable and Multivariable Logistic Regression Analysis

Baseline variable	Univariable		Multivariable*
	OR (95% CI)	P value	OR (95% CI)	P value
LVEF ≥30%	2.96 (1.55–5.65)	0.001
LVEDDi ≤35 mm/m2	3.31 (1.85–5.95)	<0.001
LVESVi ≤84 mL/m2	2.50 (1.42–4.40)	0.001
LVMi ≥202 g/m2	2.53 (1.11–5.79)	0.028
Moderate/severe FMR	0.41 (0.22–0.74)	0.003
LV‐GLS ≤−5.5%	3.07 (1.76–5.36)	<0.001	2.12 (1.10–4.10)	0.025
LV‐MD ≤76 msec	7.24 (3.97–13.21)	<0.001	5.47 (2.91–10.29)	<0.001

FMR indicates functional mitral regurgitation; GLS, global longitudinal strain; LV, left ventricular; LVEDDi, left ventricular end‐diastolic diameter index; LVEF, left ventricular ejection fraction; LVESVi, left ventricular end‐systolic volume index; LVMi, left ventricular mass index; LVRR, left ventricular reverse remodeling; MD, mechanical dispersion; and OR, odds ratio.

^*Variables included in multivariable logistic regression analysis were LVEDDi, LVESVi, LVMi, moderate/severe FMR, LV‐GLS, and LV‐MD. LVEF was excluded from the model due to the collinearity with LV‐GLS.

LV‐MD and Its Association With LVRR

The receiver operating characteristic analysis determined the optimal LV‐MD cutoff was 76 milliseconds for predicting LVRR (Figure 3). Out of the entire cohort 112 patients had LV‐MD >76 milliseconds (47%), of whom 84% (n=94) had no LBBB. The baseline characteristics of the subgroups according to LV‐MD were summarized in Table S5. Patients with LV‐MD >76 milliseconds had higher LV dimensions and volumes, lower LV‐GLS magnitude, and higher concentrations of NT‐proBNP compared with patients with LV‐MD ≤76 milliseconds. As a continuous variable, LV‐MD had a slight but significant correlation with LVEF (r=−0.192, P=0.003), LV‐GLS (r=0.353, P<0.001), and NT‐proBNP (r=0.194, P=0.003). Additionally, age, HF duration, QRS duration, and comorbidities of the patients also differed across the various LV‐MD values.

Figure 3. ROC curves of baseline LV‐MD and LV‐GLS to predict LVRR.

ROC curves showing the AUC of baseline LV‐MD (red curve) and LV‐GLS (green curve) in the prediction of LVRR. The DeLong test was used to compare the AUCs. AUC indicates area under the curve; GLS, global longitudinal strain; LV, left ventricular; LVRR, left ventricular reverse remodeling; MD, mechanical dispersion; and ROC, receiver operating characteristic.

The optimal LV‐GLS cutoff was −5.5% for predicting LVRR. Figure 3 shows that the area under the curve of LV‐MD was 0.76 (95% CI, 0.70–0.83), which was significantly higher than that of LV‐GLS (area under the curve, 0.66 [95% CI, 0.59–0.73]; P _difference=0.010). Table S6 shows the comparison of the predictive performance between LV‐GLS and LV‐MD for LVRR. Moreover, as depicted in Figure 4, the group of patients with both LV‐MD ≤76 milliseconds and LV‐GLS ≤−5.5% had the highest likelihood of LVRR (87%), followed by the group with either LV‐MD >76 milliseconds or LV‐GLS>−5.5% (58%). The group of patients with LV‐MD >76 milliseconds and LV‐GLS>−5.5% had the lowest likelihood of LVRR (31%).

Figure 4. Prevalence of LVRR stratified by LV‐GLS and LV‐MD.

Group A: LV‐MD >76 msec and LV‐GLS>−5.5%; Group B: LV‐MD >76 msec or LV‐GLS>−5.5%; Group C: LV‐MD ≤76 msec and LV‐GLS ≤−5.5%. GLS indicates global longitudinal strain; LV, left ventricular; LVRR, left ventricular reverse remodeling; and MD, mechanical dispersion.

We also determined the influence of baseline LV‐MD as a continuous variable on changes of LV function and geometry (Figure 5). The absolute increase of LVEF (r=−0.306, P<0.001), as well as relative reduction of LV end‐diastolic volume (r=0.259, P<0.001) and reduction of end‐systolic volume (r=0.311, P<0.001), were more pronounced with the decreasing LV‐MD value.

Figure 5. Relationship between baseline LV‐MD and changes in LVEF, LVEDV, and LVESV at follow‐up.

The Spearman rank correlation test was used to determine the correlation between baseline LV‐MD and changes of LVEF (A), LVEDV (B), and LVESV (C) at follow‐up. LVEF change was defined as the absolute difference between baseline and follow‐up LVEF. LVEDV and LVESV changes were calculated as absolute differences divided by absolute values at baseline. LV indicates left ventricular; LVEDV, left ventricular end‐diastolic volume; LVEF, left ventricular ejection fraction; LVESV, left ventricular end‐systolic volume; LVRR, left ventricular reverse remodeling; and MD, mechanical dispersion.

Additive Value of LV‐MD for Predicting LVRR

Three predictive models for LVRR, including the clinical model (model 1) that integrated only clinical predictors, the model 2 with LV‐GLS addition, and the model 3 with both LV‐GLS and LV‐MD addition, were constructed. After adjustment for clinical parameters and LV‐GLS, LV‐MD ≤76 milliseconds remained a significant predictor of LVRR (OR 7.61 95% CI 3.39–17.08, P<0.001). Sensitivity analysis incorporating the multiply imputed data from patients lost to follow‐up into the primary analysis population yielded consistent results (adjusted OR, 4.43 [95% CI, 1.89–10.36], P=0.002).

The comparison of their prognostic value for LVRR was exhibited in Figure 6 and Table S7. The model 1 achieved a C‐statistic of 0.82 for LVRR prediction and addition of LV‐GLS ≤−5.5% improved global model fit (chi‐square: 77.0 versus 70.6, P=0.011), but not discrimination (C‐statistic: 0.83 versus 0.82, P=0.368). Interestingly, on top of the model 2, addition of LV‐MD ≤76 milliseconds further improved global model fit (chi‐square: 105.0 versus 77.0, P<0.001) with better discrimination (C‐statistic: 0.88 versus 0.83, P=0.009). Figure S2 illustrates the calibration curves for all 3 models. By 10‐fold cross‐validation, the final model demonstrated consistent performance across training and validation sets. The area under the curve of the training set was 0.87±0.01, closely aligning with the area under the curve of 0.85±0.06 in the validation set. The minimal difference (2%) suggested robust generalization without overfitting.

Figure 6. Incremental value of MD over clinical parameters and GLS as a correlate of LVRR.

*P value according to comparison between chi‐square values of different models performed by likelihood ratio test. ARNI indicates angiotensin receptor‐neprilysin inhibitors; GLS, global longitudinal strain; HF, heart failure; LV, left ventricular; LVRR, left ventricular reverse remodeling; MD, mechanical dispersion; OR, odds ratio; and SBP, systolic blood pressure.

Moreover, the association of LV‐MD with LVRR in different clinical settings was also evaluated. LV‐MD remained an independent predictor of LVRR in subgroups stratified by age, sex, and HF cause. Exclusion of patients with LBBB or CRT did not change the association between LV‐MD and LVRR (Table S8).

LV‐MD During Follow‐Up

At 1‐month follow‐up, LV‐GLS significantly improved (−8.3% versus −6.5%, P<0.001), whereas LV‐MD did not (76 milliseconds versus 73 milliseconds, P=0.108) (Table S9). As shown in Figure S3, the proportion of LV‐MD ≤76 milliseconds increased from 53% at baseline to 66% at the end of our study (P=0.004). The 57 patients with baseline LV‐MD >76 milliseconds had LV‐MD ≤76 milliseconds at a median follow‐up of 7 months (IQR, 1–13 months). The proportion of LV‐MD ≤76 milliseconds significantly increased during follow‐up in patients with LVRR (84% versus 70%, P=0.003), but not in the group without LVRR (34% versus 24%, P=0.134).

Reproducibility

Both LV‐MD and LV‐GLS showed good intraobserver (ICC, 0.87 [95% CI, 0.67–0.95], P<0.001, and ICC, 0.97 [95% CI, 0.94–0.99], P<0.001, respectively) and interobserver agreement (ICC, 0.85 [95% CI, 0.67–0.94], P<0.001, and ICC, 0.95 [95% CI, 0.82–0.98], P<0.001, respectively).

DISCUSSION

To the best of our knowledge, the present study is the first to investigate the correlation between LV‐MD, an echocardiographic indicator of LV mechanical dyssynchrony, and LVRR in HFrEF. The principal findings were that (1) LV mechanical dyssynchrony was common in our cohort with HFrEF, regardless of the presence of LBBB; (2) LV‐MD was significantly associated with LVRR, independent of and superior to LV‐GLS; and (3) patients with remarkably prolonged LV‐MD (>76 milliseconds) and impaired LV‐GLS (>−5.5%) on top of clinical predictors had the lowest likelihood of LVRR.

During a median follow‐up period of 19 months, 64% of our patients with HFrEF achieved LVRR, which was higher compared with previous investigations showing that LVRR occurred in 30% to 50% of patients. ⁴ , ⁵ , ⁶ In our study population, 67% and 36% of all patients received ARNIs and SGLT2is, whereas these novel medicines were either not used or used at lower frequencies in earlier studies. Younger age, shorter HF duration, higher SBP, narrow QRS duration, and ARNI use were associated with LVRR, which was in keeping with former studies. ⁴ , ⁶ , ²³ , ²⁴ In contrast, NT‐proBNP, HF cause,and severity of functional mitral regurgitation identified in previous studies ² , ²⁵ were significant only in univariable analysis but not multivariable analysis in our study, which remains to be further investigated.

Imaging parameters, such as late gadolinium enhancement on CMR, has been proven to be an independent predictor of LVRR. ²⁶ However, CMR cannot be routinely used. Other widespread available parameters by echocardiography, such as LVEF or chamber dimensions, did not show a consistent association with LVRR. ⁶ , ⁷ In comparison, speckle‐tracking echocardiography‐derived LV‐GLS is more sensitive than LVEF in detecting subclinical cardiac dysfunction, ¹¹ and studies have demonstrated that LV‐GLS was a correlate of mortality, independent of and incremental to LVEF in HFrEF. ¹⁰ , ¹¹ , ¹² Swat et al ⁹ further showed that higher baseline LV‐GLS (≤−8%) was associated with recovery of LVEF in patients with nonischemic cardiomyopathy with reduced LVEF. Another retrospective study by Jung et al ²⁷ also found that LV‐GLS was an independent predictor for LVRR in dilated cardiomyopathy, with the optimal LV‐GLS cutoff of −10%. In the present study, we confirmed the adverse impact of impaired LV‐GLS on LVRR in HFrEF, and an optimal LV‐GLS threshold of −5.5% was here identified, the magnitude of which was lower than those in previous reports. ⁹ , ²⁷ The lower cutoff in our cohort might be due to the differences in cause, severity, and therapeutic choices of HF among studies. Compared with previous studies, ⁹ , ²⁷ our cohort included patients with both ischemic and nonischemic causes, was more extensively treated with ARNI, and had more dilated LV chamber at baseline.

LV‐MD is another strain‐derived parameter, which represents the dyssynchrony of segmental myocardial contraction across cardiac cycles. Although its normal reference remains to be determined, 2 studies recently reported the upper limit of normality of LV‐MD, being 61 milliseconds in healthy middle‐aged and 64 milliseconds in healthy older subjects, respectively. ²⁸ , ²⁹ LV mechanical dyssynchrony, reflected in prolonged LV‐MD, has been reported to be present and associated with worse outcomes in many cardiovascular diseases, including cardiomyopathy and myocardial infarction, and LV‐MD of 60 milliseconds to 75 milliseconds could predict ventricular tachycardia and sudden cardiac death in cardiomyopathy. ¹³ , ¹⁴ , ¹⁵ In our cohort with HFrEF, baseline LV‐MD was also remarkably lower in patients who were alive after 12 months than those who died within 12 months. Besides, LV‐MD ≥68 milliseconds was significantly associated with an increased risk of mortality in severe aortic stenosis. ³⁰ In HF, echocardiographic parameters reflecting dyssynchrony, such as the presence of septal flash and apical rocking, time differences based on radial strain and patterns of regional longitudinal strain, have been linked to LVRR in patients receiving CRT treatment. ¹⁸ , ¹⁹ , ²⁰ A recent study reported that LV‐MD was a predictor of CRT response with an optimal cutoff value of 106 milliseconds. ¹⁷ However, whether LV‐MD had an impact on LVRR in general patients with HF was unknown.

In the present study, 94 patients (84%) with LV‐MD >76 milliseconds had no LBBB, suggesting that LV mechanical dyssynchrony was widely present and irrespective of the presence of LBBB in HFrEF. Moreover, a great heterogeneity of LV‐MD was seen, with the LV‐MD ranging from 30 to 202 milliseconds, which influenced by mixed factors including age, HF duration and severity, QRS duration, and comorbidities. The median LV‐MD in our cohort was 73 milliseconds, which was much higher than that in healthy controls. ²⁸ , ²⁹ In contrast, a narrower heterogeneity of LV‐MD (IQR, 49–77 milliseconds) and a lower median value of LV‐MD (61 milliseconds) were reported in aortic stenosis. ³⁰

With respect to the potential role of LV‐MD in predicting LVRR, we found that LV‐MD was significantly correlated with the presence and magnitude of LVRR. The optimal LV‐MD cutoff value for predicting LVRR in our study was 76 milliseconds, which was notably lower than that reported in previous study involving solely CRT recipients (106 milliseconds). ¹⁷ The possible explanation was that CRT candidates had a greater degree of dyssynchrony and thus higher LV‐MD. After adjustment for clinical and echocardiographic parameters, including LV‐GLS, the prognostic value of LV‐MD for LVRR remained significant. Importantly, exclusion of patients with LBBB did not alter the results. In comparison with LV‐GLS, our receiver operating characteristic analysis suggested that LV‐MD was superior to LV‐GLS in predicting LVRR.

Why LV‐MD is able to predict LVRR is likely ascribed to several aspects. LV‐MD is associated with HF severity, as suggested by our findings that it had a weak correlation with LVEF, LV‐GLS, and NT‐proBNP. Furthermore, LV‐MD may provide information beyond traditional measurements in cardiac structure and function. Patients with HF with prolonged LV‐MD exhibit greater CMR‐defined myocardial fibrosis or ischemia, which may induce larger electromechanical LV activation delay, ³¹ , ³² and are less likely to have LVRR. Additionally, timing of peak regional deformation also reflects nonuniform loading conditions in a diseased ventricle that are related to LVRR as well. ¹¹ A previous study demonstrated that early improvement of LV‐GLS at 3 months was associated with significant LVEF improvement at 12 months. ³³ Similarly, another study found that reduced LV‐MD at 6 months after CRT was correlated with lower risk of mortality and ventricular arrhythmia. ³⁴ Therefore, we additionally observed the recovery speed of LV‐GLS and LV‐MD during follow‐up. In our study, at 1‐month follow‐up, LV‐GLS improved significantly, but LV‐MD did not. It seemed that improvement of LV‐GLS preceded that of LV‐MD and LVRR in our study. Further studies are needed to investigate the temporal changes of LV‐MD, and importantly, its influence on LVRR.

Because LV‐MD together with LV‐GLS provided combined information on LV mechanical dyssynchrony and cardiac function, we designed a novel LVRR predictive model by adding both LV‐MD ≤76 milliseconds and LV‐GLS ≤−5.5% to a model integrating age, HF duration, SBP, QRS duration, and ARNI use and hypothesized that it may improve the predictive power. As a result, this novel model greatly improved the predictive performance in terms of LVRR. Reliable prediction of both forward and reverse remodeling facilitates risk stratification and is pivotal for clinical decision‐making process and tailored management in HF. In patients at higher risk for persistent LV dilation and function decline, closer follow‐up and more rapid escalation of therapy would be necessary. On the contrary, in patients who are prone to experience LVRR, waiting may be warranted to avoid untimely device implantation or heart transplantation.

Study Limitations

This study has several limitations. First, this was a single‐center observational study with a relatively limited number of patients. The patients enrolled in the present study were those who were referred to our tertiary care center and likely with more severe disease. The proportion of loss to follow‐up is relatively high, though the patients included in the analysis were roughly compatible with those lost to follow‐up. Therefore, selection bias is likely present. Second, the predictive value of LV‐MD was significant in patients with either ischemic or nonischemic cause of HF. However, whether the predictive role of LV‐MD varies with HF cause cannot be answered by the study and further studies are needed. Third, LVRR could occur from as early as 3 months to up to 2 years after initiation of treatments ⁷ and thus the median follow‐up of 19 months in the study may be inadequate. However, LVRR occurred in 152 patients (64%) at a median time of 7 months (IQR, 6–13 months) in our study and it was reasonable to presume that the follow‐up period of the study was relatively adequate for the majority of our patients to achieve LVRR. Fourth, myocardial fibrosis by CMR is a recognized predictor of LVRR and our study did not include data of CMR. Finally, due to short time to market in China, SGLT2is that have been demonstrated to improve HF outcomes and cause a great reduction in LV mass ³⁵ were used in only 36% of our patients.

CONCLUSIONS

LV mechanical dyssynchrony is widely present in HFrEF, even in patients without LBBB. Echocardiographic strain‐derived LV‐MD is an independent predictor of LVRR and provides additional prognostic value over LV‐GLS. Patients with both prolonged LV‐MD and impaired LV‐GLS have lower likelihood of LVRR. Addition of LV‐GLS and LV‐MD to traditional clinical parameters, including age, HF duration, SBP, QRS duration, and ARNI use, results in improvement for LVRR prediction and risk stratification in patients with HFrEF.

Sources of Funding

None.

Disclosures

None.

Supporting information

Tables S1–S9

Figures S1–S3