1. Introduction
Sudden cardiac death (SCD) accounts for approximately 25% of cardiovascular death, and implantable cardioverter defibrillator (ICD) effectively reduces mortality related to sudden death. Current ventricular arrhythmias (VA) and SCD guidelines recommend primary ICD implantation in symptomatic (New York Heart Association (NYHA) function class II-III) ischemic cardiomyopathy (ICM)/non-ischemic cardiomyopathy (NICM) patients and left ventricular ejection fraction (LVEF) ≤35% after ≥3 months of optimal medical treatment [
1,
2]. LVEF is the core indicator of risk stratification for patients with a risk of SCD, and the cut-off of 35% is derived from the pivotal clinical trials conducted approximately 20 years ago [
3,
4,
5].
Other echocardiographic characteristics such as LV mass, left atrial dimension and left ventricular end-systolic dimension have been shown to be associated with clinical outcomes [
6]. Among these, baseline left ventricular end-diastolic diameter (LVEDD) has been associated with all-cause mortality and SCD in the general population and heart failure patients [
7,
8,
9,
10,
11]. However, the impact of LVEF and LVEDD on all-cause mortality and SCD has not been thoroughly investigated in patients with ICD.
Therefore, the aim of this study is to quantify the prognostic value of LVEF and LVEDD for clinical outcomes, respectively. We also investigated the incremental value of adding LVEF and LVEDD to the established clinical risk predictors.
2. Methods
2.1. Population
We retrospectively included patients from a maintained ICD database at Fuwai Hospital from March 2010 to September 2020. Eligible patients were age older than 18, had complete echocardiography examination and had a diagnosis of ICM or NICM. For the indication of ICD, primary prevention was defined as in symptomatic ICM/NICM patients with NYHA function class II-III and LVEF ≤ 35% following optimal medical treatment for at least 3 months. The secondary prevention was defined as prophylactic ICD implantation to prevent sudden cardiac death in patients with a history of life-threatening ventricular arrhythmia [
1,
2]. ICM was defined as significant stenosis (≥50% of left main or ≥70% of major epicardial coronary artery) on coronary angiography or with evidence of myocardial infarction. We excluded patients who were lost of follow-up, without current implantation indication of ICD [
1], or who had undergone device replacement. This study was conducted in accordance with the Declaration of Helsinki and approved by the institutional review board and ethics committees. All patients provided written informed consent.
2.2. Echocardiography
Electrocardiogram-triggered two-dimensional echocardiographic examination was performed before the implantation of ICD by experienced sonographers using commercially available equipment (GE Vivid) and interpreted by experienced cardiologists. LVEF was calculated using the biplane summation-of-disks method (modified Simpson’s rule) in the apical four- and two- chamber views. LVEDD was measured at a level just below the mitral valve leaflet tips in the parasternal long-axis view [
12].
2.3. Follow-Up and Outcomes
ICD patients underwent routine interrogation and follow up at Fuwai hospital. The primary endpoint of this study was all-cause mortality. The secondary endpoint was the occurrence of appropriate ICD shock as a substitute for SCD. Appropriate ICD shock was defined as shocks for ventricular tachycardia (VT) or ventricular fibrillation (VF). Because this study included patients with secondary prevention, the ICD programming parameters for VT and VF varied among patients.
2.4. Statistical Analysis
Continuous variables were presented as means and standard deviation (SD) or medians and interquartile range (IQR) as appropriate. Categorical variables were presented as counts and percentages. Differences among quintiles were analyzed using one-way ANOVA or Kruskal-Wallis test for continuous variables and Pearson chi-square test or Fisher’s exact test for categorical variables. For variables missing less than 5%, they were considered to be missing at random, and random forest imputation was used. The correlation between potential predictors was performed by calculating Pearson’s correlation coefficient. When predictors demonstrated a high degree of collinearity, only one factor was considered to be included in the regression model. Multivariable Cox regressions were used to investigate the relationship of LVEF and LVEDD to all-cause mortality. Fine-Gray analyses accounting for all-cause mortality as a competing risk were used to investigate their relationships to appropriate ICD shocks. We adjusted models for important clinical predictors as well as for covariates that were significant in univariable analysis. For the Cox regression analysis, our basic model (model 1) started with several clinical predictors based on prior research and univariable Cox regression results: age, sex, BMI, systolic blood pressure, NYHA class >II and indication of ICD. Model 2 was additionally adjusted for comorbidities, including complete right bundle branch block (RBBB), atrial fibrillation (AF), diabetes, and ICM. Model 3 was additionally adjusted for medications including digoxin, diuretic and ACEI/ARB/ARNI compared with model 2. Model 4 further included eGFR, albumin and NT-proBNP. For the analysis of the secondary endpoint, model 1 adjusted for covariates mentioned above in Cox regression model 1, as well as for smoking and alcohol. The additional variables in models 2– 4 were the same as in the Cox regression model.
To evaluate the association of LVEF and LVEDD with the endpoints, we first stratified the population into quintiles according to LVEF and LVEDD, respectively. Then P for linear trend was calculated. The potential non-linearity was tested using penalized spline terms and the association between the echocardiographic parameter and the risk of the endpoint was plotted using penalized splines. Secondly, we treated the echocardiographic parameter continuously if it did not violate the linearity assumption of the Cox regression.
The performance of models was compared using C-statistic improvement. The likelihood ratio test was used to evaluate the fit of the models. Category-free net reclassification index (NRI) and integrated discrimination index (IDI) were assessed to measure the prediction improvement of LVEF and LVEDD. Possible interactions of LVEF or LVEDD with sex, the indication of ICD, ICM, and CLBBB were tested. Furthermore, we performed sensitivity analyses examining the robustness of the results using complete-case analysis. Statistical analysis was performed in R version 4.1.2 (Vienna, Austria). Multivariable Cox regressions were performed using the survival (version 3.2-13) package and competing risk models were fitted using the cmprsk (version 2.2-10) package.
4. Discussion
The main findings of our study including patients with ICD implantation are as follows: (1) patients with lower LVEF or larger LVEDD had a higher risk of all-cause mortality during follow-up. (2) patients with larger LVEDD at baseline had a higher risk of appropriate shock. (3) Moreover, the addition of LVEDD to clinical models provided incremental prognostic information over LVEF for both endpoints. These findings indicate that LVEDD could be a helpful predictor of all-cause mortality and appropriate shock better than LVEF in patients with ICD implantation.
According to current guidelines, LVEF is the core indicator of the selection criterion for primary prevention [
2]. LVEF was a strong, independent predictor of SCD and cardiovascular mortality in pivotal trials [
5,
13]. For HFrEF patients, ICD was associated with 23% relative risk reduction in all-cause mortality in pivotal RCTs [
14]. However, the DANISH trial in which the inclusion criteria required LVEF ≤ 35%, failed to show the survival benefit of ICD in NICM patients [
15]. Prior studies have shown that LVEF is neither highly sensitive nor specific in regard to SCD prediction [
16,
17]. New predictors for risk prediction of clinical outcomes are urgently warranted. A suitable marker for patients should therefore predict not only all-cause mortality but also the risk of arrhythmic risk. Our study shows that LVEDD could be a valuable marker to predict both clinical outcomes. Moreover, LVEF cutoff of 35% bypassed the rigorous steps of evaluation to test whether it is sufficient to predict clinical outcomes [
17]. Our results provide the first detailed, quantified relationship between echocardiographic parameters (LVEF and LVEDD) and clinical outcomes. Our results suggest a linear relationship between LVEF and LVEDD to all-cause mortality; however, for the appropriate shocks, a linear association was only identified with LVEDD.
Several observational studies demonstrated that LVEDD was associated with the risk of all- cause mortality and ventricular arrhythmias in patients with ICD [
7,
18,
19]. Sataka et al. showed that LVEDD ≥ 65 mm was an independent predictor of fatal arrhythmic events in patients with primary prevention for class I and class IIa indication in Japan [
20]. Another study including 853 patients with only Biotronik ICD demonstrated that moderate LVEF dysfunction (35–45%) and left ventricular enlargement (LVEDD ≥ 60 mm) were associated with a higher risk of VA and cardiac death [
21]. Moreover, in a sub-study of the Genetic Risk of Assessment of Defibrillator Events (GRADE) study, LVEDD was associated with appropriate shocks (HR for per 10 mm: 1.33,
p < 0.001) and the composite endpoint of death, transplant and left ventricular assist device (HR for per 10 mm 1.40
p < 0.001) [
7]. Our study supports these results. However, LVEF was found to be both associated with shocks (HR for per 1% 0.95,
p = 0.0002) and the composite endpoint (HR for per 1% 0.95,
p < 0.001) in this study [
7]. This is different from our results. Of note, there were several key differences between the two studies. First, this study ignored the competing risk of death. Therefore, this could overestimate the risk of appropriate shocks and might distort the results [
22]. Second, the differences in patients’ characteristics might contribute to the controversial results. The inclusion criterion was LVEF < 30%, and the majority of patients (74.61%) had the primary indication in the sub-study of the GRADE study. However, in our study, the mean LVEF was 37.3 ± 11.7% and 23.1% of patients had the primary indication. Subgroup analyses in our study indicated that there was no interaction between the predictive value of echocardiographic parameters and the indication of ICD.
Larger LVEDD represents dilation of the left ventricular. Left ventricular dilation reflects the alternations in the expression of connexins and stretch-mediated changes in ion channel function and adverse left ventricular remodeling, and could create a favorable milieu for re-entrant ventricular arrhythmias [
23,
24]. In animal models of dyssynchronous heart failure, LV end- diastolic volume was significantly larger in the heart failure (HF) model. Moreover, remodeling of K
+ current and Ca
2+ current and prolongation of action potential duration (APD) were shown in the dyssynchronous HF model, which might be cellular triggers and substrate for arrhythmias [
23,
25]. Another mechanism underlying these might be attributable to myocardial fibrosis. Prior study has demonstrated that NICM patients with fibrosis had higher left ventricular end-diastolic volumes and the myocardial fibrosis was independently associated with all-cause mortality and SCD [
26,
27].
In this study, calculation of LVEF and measurement of LVEDD were according to current recommendations [
12]. However, the accuracy of LVEF calculation was in excess of ±10% for Simpson’s rule with echocardiography [
28]. Thus, LVEF might have limited reproducibility, especially in 2D imaging, which is commonly used in clinical practice. In contrast, LVEDD was measured using an electronic caliper and a line at the certain level just below the mitral valve leaflet tips, which enhanced the reproducibility of the results.
For the optimal cut-off of LVEDD, previous studies applied different cut-off values according to the characteristics of patients. In CHART-1 study, LVEDD > 60 mm was associated with SCD in heart failure patients, while in CHART-2 study, LVEDD > 65 mm rather than 60 mm was a significant risk factor [
8,
20]. In our study, we decided not to define the optimal cut-off using the Youden index in receiver-operating characteristic, as this could lead to a type-I error inflation. Moreover, LVEDD represents a continuum of risk, and the aim of this study is to qualify the relationship of LVEDD and clinical outcomes. Thus, we moved from a dichotomous to a continuous risk stratification.
LVEDD and LVEF in this study were measured at baseline. Prior study has proven that 24% of NICM patients with prophylactic ICD implantation had left ventricular function improvement (LVEF increase ≥5% and LVEDD decrease ≥5 mm) within 1 year [
29]. However, our study shows that LVEDD was still associated with clinical outcomes for almost ten years of follow-up, and future studies are warranted to assess the value of dynamic examination of LVEDD.
Limitations
Our study has several limitations. First, the main limitation of the present study is its retrospective observation design. However, it reflects the real-world clinical practice and prospective research is needed in the future. Second, in our study, the appropriate shock was the secondary outcome as the substitute for SCD. We acknowledge that SCD is a more meaningful endpoint. Moreover, a study showed that ICD therapies are more frequent than sudden death, with approximately twice as many shocks as fatal arrhythmias [
30]. However, we did not obtain the causes of death, thus we could only use appropriate shock as a surrogate for SCD. Third, echocardiography was measured at baseline, and the association of temporal change of LVEDD and clinical outcomes was not evaluated. In our study, LVEDD was still associated with clinical outcomes after 59.6 months of follow-up. Future studies are needed to assess the relationship between LVEDD temporal change and clinical outcomes. Fourth, we included a mixed population of patients with ischemic and non- ischemic cardiomyopathy and primary and secondary prevention indication, while there were no significant interactions in subgroup analysis. Fifth, we did not have the genetic testing results to perform a subgroup analysis. Finally, as granular CMR data were not available in this study, no further relationship between LVEDD and fibrosis could be investigated.