Myocardial strain assessment using cardiovascular magnetic resonance imaging in recipients of implantable cardioverter defibrillators

The implantable cardioverter-defibrillator (ICD) reduces mortality in patients at high risk of sudden cardiac death. Randomized clinical trials and observational studies have established a role for ICD in patients with left ventricular (LV) systolic dysfunction and heart failure [1, 2], the prevention of recurrent cardiac arrest [3‐5], and inherited conditions that predispose to sudden death [6]. However, many patients who meet the guideline-recommended criteria for ICD implantation never receive appropriate therapy for ventricular arrhythmias [7], and further risk stratification would be helpful.

Myocardial strain imaging has emerged as a useful clinical tool that can provide incremental information about LV systolic function beyond global assessment by LV ejection fraction (LVEF). LV strain imaging is a validated method of assessing regional and global function with demonstrable prognostic value in multiple settings, including post-myocardial infarction [8] and heart failure [9]. Feature tracking (FT) cardiovascular magnetic resonance imaging (CMR) is a recently developed method to derive strain and wall mechanics, which allows tracking of tissue voxel motion on cine CMR images [10]. Assessment of global longitudinal strain (GLS) and global circumferential strain (GCS) by FT-CMR has been shown to be well-correlated with strain assessed by CMR myocardial tagging [11, 12]. Whilst CMR tagging remains the gold standard for CMR-derived strain, it requires dedicated imaging sequences. Both methods remain generally comparable to speckle-tracking echocardiography but without the field of view limitations [13].

Despite the established prognostic value of CMR-derived strain, there are limited data on its utility for predicting adverse outcomes following ICD implant. Accordingly, the objective of our study was to examine the predictive value of LV strain parameters as measured by FT-CMR on outcomes following primary and secondary prophylactic ICD placement.

A total of 364 patients were studied, and their baseline characteristics are presented in Table 1. The mean age of the cohort was 61 years and 79% were male; 61% had an ICD implanted for a primary prevention indication, and 57% had ischemic heart disease. The median time from CMR to ICD implant was 29 days. Over a median follow-up of 62 months (IQR: 39–101), 75 patients (20.6%) had appropriate ICD shocks and 64 patients (17.6%) died; 118 (32.4%) of the cohort reached the primary composite endpoint. Correlations between strain parameters and LVEDVI, LVESVI, LVMI and LVEF are summarized in Table 2. There were negative correlations between LVEF and GLS (lower LVEF correlated with less negative GLS value) and GCS (lower LVEF correlated with less negative GCS value), and positive correlation between LVEF and GRS (lower LVEF correlated with less positive GRS value).

Figure 1 illustrates the Kaplan–Meier curves of the primary composite endpoint for each strain parameter by quartiles. Significant associations with appropriate ICD shocks or death were demonstrated for GLS (Fig. 1A), GCS (Fig. 1B), and GRS (Fig. 1C) (all p for trend < 0.01). Given the collinearity between strain parameters and LVEF, multivariable analyses excluding LVEF but incorporating age, indication (primary versus secondary prevention), presence of ischemic heart disease, and presence of LGE demonstrated GLS, GCS, and GRS to be independently predictive of both the primary outcome and all-cause mortality (Table 3). When the analyses were stratified by primary versus secondary prevention, the results were similar, and there was no significant interaction (all p for interaction > 0.10). In the multivariable analyses incorporating LVEF, GLS remained a significant independent predictor of the primary endpoint (HR 1.05 per 1% higher GLS, 95% CI 1.01–1.09, p = 0.010) and all-cause mortality (HR 1.06 per 1% higher GLS, 95% CI 1.02–1.11, p = 0.003). With respect to model discrimination, Harrell’s c were 0.600 and 0.613, and Somers’ d were 0.200 and 0.225, respectively, for the models without and with GLS for the primary endpoint; Akaike’s information criterion was lower (i.e., better) for the model including GLS. In a sensitivity analysis adjusting for quantitative LGE by visual assessment, the findings were unchanged (HR 1.05 per 1% higher GLS, 95% CI 1.01–1.09, p = 0.01). GLS was also independently associated with the composite endpoint of appropriate anti-tachycardia pacing, ICD shock or death (HR 1.08 per 1% higher GLS, 95% CI 1.04–1.12, p < 0.001), which was maintained (HR 1.06 per 1% higher GLS, 95% CI 1.01–1.11, p = 0.020) after further adjustment for LVEF.

Multivariable Cox regression using restricted cubic splines for each strain parameter did not show clear departure from linearity for GLS, GCS, and GRS. There was no significant interaction between GLS and a primary prevention indication for ICD implant (p = 0.32), ischemic heart disease (p = 0.92), presence of LGE (p = 0.86), or gender (p = 0.32).

Intra-observer variability was excellent, with intraclass correlation coefficients for absolute agreement of 0.94 (95% CI 0.85–0.98) for GLS, 0.95 (95% CI 0.83–0.98) for GCS, and 0.95 (95% CI 0.86–0.98) for GRS (all p < 0.001).

In this study of primary and secondary prevention ICD patients with ischemic and non-ischemic cardiomyopathy, CMR-derived LV strain parameters were significantly correlated with LV volumes, LVM, and LVEF. In both univariate and multivariable analyses excluding LVEF for collinearity, reduced GLS, GCS, and GRS values were associated with appropriate ICD shocks and death, as well as all-cause mortality alone. Furthermore, GLS remained a predictor of appropriate ICD shocks and mortality independently of LVEF, regardless of whether the ICD indication was primary or secondary prevention, the presence of ischemic heart disease, or the presence of scar/infarct.

It is recognized that LVEF alone has limited discrimination for predicting sudden death [18]. CMR has an established role in improving risk stratification through improved tissue characterization, specifically focused on both the patterns and quantification of LGE in ischemic and nonischemic cardiomyopathies alike [19, 20]. Although speckle tracking echocardiography derived GLS measurements have demonstrated prognostic value, they are dependent on high quality images. Our study utilized a CMR parameter that does not require additional image acquisition, is rapidly acquired and feasible on almost all routine CMR cine images, and therefore could potentially be used to refine ICD patient selection. As the availability of CMR increases—providing accurate LVEF and right ventricular ejection fraction (RVEF) assessment [21‐23], scar quantification, tissue characterization [24], assessment of myocardial viability [25], quantitative valvular assessment and now, LV strain analysis—its role in the noninvasive evaluation of arrhythmia risk will become even more integral.

Myocardial strain is a dimensionless entity, referring to the change in length of myocardium relative to its original length. Subendocardial myocardial fibres are responsible for long-axis contraction, whereas mid-myocardial and sub-epicardial fibres are responsible for circumferential and radial contraction. GLS has demonstrated utility, independent of LVEF, as a sensitive parameter of LV function predictive of clinical outcomes across a wide spectrum of cardiac diseases, including myocardial infarction, heart failure, detection and monitoring of chemotherapy cardiotoxicity, and valvular heart disease [26, 27]. In regards to its ability to predict adverse arrhythmic outcomes, it is conceivable that anatomic abnormalities that predispose to arrhythmia (for example, due to heterogeneity in conduction through myocardial scar) also result in heterogeneous regional contraction, which in turn is reflected in strain parameters. LV strain is a less load-dependent measure of contractile function than LVEF and a more sensitive measure of impaired contractility. Furthermore, derangements in regional contractility, induced by diffuse, as opposed to focal myocardial fibrosis, may be identified using FT-CMR derived strain. Indeed, echocardiography-derived mechanical dispersion (a parameter determined from myocardial strain measurements) has been demonstrated to have incremental predictive value, independent of LVEF, for ventricular arrhythmias [28].

Methods of assessing strain by CMR include tissue phase mapping, strain-encoded imaging, myocardial tissue tagging, displacement encoding with simulated echoes, and FT-bSSFP imaging [29]. Although advances in myocardial tissue tagging with methods such as harmonic phase imaging and strain-encoded imaging have significantly reduced analysis time, acquisition and post-processing of additional sequences are still required. In contrast, FT-CMR uses semi-automated volumetric measures on routinely acquired cine bSSFP images. It is therefore available across different CMR vendors and field strengths, and does not require any image conversion to alternative formats or dedicated pulse sequences. Although segmental strain has been shown to be heterogeneous with FT-CMR [10, 30], global strain parameters appear to be more robust and reproducible.

Our findings add to existing evidence supporting the prognostic significance of CMR-derived strain measurements. Romano et al. examined 1012 patients with LV dysfunction (LVEF < 50%) using FT-CMR derived GLS for predicting all-cause mortality, and determined that it afforded incremental prognostication for all-cause mortality beyond LVEF and LGE in both ischemic and non-ischemic cardiomyopathy subgroups [31]. Mordi et al. found GCS derived by harmonic phase myocardial tagging to be independently prognostic in their single-centre study of 539 unselected patients undergoing CMR after adjusting for LVEF and LGE [32]. Paiman et al. found the extent of moderately impaired LV GCS to be associated with appropriate ICD therapy in 121 patients with primary prevention ICDs and history of previous myocardial infarction [33]. However, this original study was relatively small with only 30 events and, the endpoint of appropriate ICD therapy included ATP in addition to shocks. Furthermore, other global strain parameters were not assessed, and patients with non-ischemic cardiomyopathy were not included. Our study extends the prognostic value of LV strain to ICD recipients, and may have a future role in directing primary prevention ICD treatment to patients more likely to derive benefit than those selected based only on current risk stratification models which are predominantly influenced by LVEF.

We selected a primary composite endpoint encompassing arrhythmia (appropriate ICD shock) and all-cause mortality with the aim of determining whether strain could help refine ICD treatment benefit. It must be acknowledged that the threshold for an ICD shock is, to some degree, dependent upon device programming, and that not all ICD shocks, even when deemed appropriate, are life-saving [7]. However, adjudicated appropriate shocks remain the most robust endpoint besides death for determining that patients derived survival benefit from the ICD. While patients with ICDs are also at elevated risk of non-arrhythmic death [34], the cause of death can be particularly difficult to ascertain in both randomized and observational studies, despite central adjudication. Our study is consistent with randomized trials of ICDs that have used all-cause, rather than disease-specific, mortality.

We had a relatively long follow-up period in this study with > 100 events, and our cohort had relatively few exclusion criteria, enhancing the generalizability of our results. All CMR measurements were performed independently by readers blinded to clinical outcomes. We focused on appropriate ICD shocks that were independently adjudicated, a more robust endpoint for survival than including either any shock or anti-tachycardia pacing [35, 36]. Similarly, we selected all-cause mortality over disease-specific mortality, consistent with the landmark trials evaluating ICDs for primary or secondary prevention, as the more clinically relevant endpoint.

Among patients who underwent ICD implant for primary and secondary prevention of sudden cardiac death, FT-CMR GLS assessment was independently predictive of appropriate shocks and all-cause mortality, even after adjustment for LVEF. These findings support the potential use of FT-CMR derived strain analyses to enhance the selection of patients most likely to benefit from ICD implant. Further studies are needed to confirm these findings in a broader spectrum of ICD candidates.