Comprehensive Phenotypic Characterization of Late Gadolinium Enhancement Predicts Sudden Cardiac Death in Coronary Artery Disease

Background Late gadolinium enhancement (LGE) cardiac magnetic resonance (CMR) offers the potential to noninvasively characterize the phenotypic substrate for sudden cardiac death (SCD). Objectives The authors assessed the utility of infarct characterization by CMR, including scar microstructure analysis, to predict SCD in patients with coronary artery disease (CAD). Methods Patients with stable CAD were prospectively recruited into a CMR registry. LGE quantification of core infarction and the peri-infarct zone (PIZ) was performed alongside computational image analysis to extract morphologic and texture scar microstructure features. The primary outcome was SCD or aborted SCD. Results Of 437 patients (mean age: 64 years; mean left ventricular ejection fraction [LVEF]: 47%) followed for a median of 6.3 years, 49 patients (11.2%) experienced the primary outcome. On multivariable analysis, PIZ mass and core infarct mass were independently associated with the primary outcome (per gram: HR: 1.07 [95% CI: 1.02-1.12]; P = 0.002 and HR: 1.03 [95% CI: 1.01-1.05]; P = 0.01, respectively), and the addition of both parameters improved discrimination of the model (Harrell’s C-statistic: 0.64-0.79). PIZ mass, however, did not provide incremental prognostic value over core infarct mass based on Harrell’s C-statistic or risk reclassification analysis. Severely reduced LVEF did not predict the primary endpoint after adjustment for scar mass. On scar microstructure analysis, the number of LGE islands in addition to scar transmurality, radiality, interface area, and entropy were all associated with the primary outcome after adjustment for severely reduced LVEF and New York Heart Association functional class of >1. No scar microstructure feature remained associated with the primary endpoint when PIZ mass and core infarct mass were added to the regression models. Conclusions Comprehensive LGE characterization independently predicted SCD risk beyond conventional predictors used in implantable cardioverter-defibrillator (ICD) insertion guidelines. These results signify the potential for a more personalized approach to determining ICD candidacy in CAD.

C urrent tools to identify patients at high risk of sudden cardiac death (SCD) are limited.
Specifically, left ventricular ejection fraction (LVEF) is an imprecise metric, and innovative approaches are required to identify arrhythmogenic substrate beyond this measure. 1 SCD risk prediction is of notable importance for patients with coronary artery disease (CAD) because their underlying etiology alone confers an enhanced risk profile. 2 It is therefore crucial to evaluate the utility of novel prediction tools to identify high-risk patients within this cohort.
In patients with chronic CAD, re-entrant ventricular tachycardia (VT) is the presumed mechanism underpinning the majority of SCD cases. 3 Septa of replacement extracellular fibrosis (resultant from necrosing myocytes) perforating bundles of surviving myocytes can provide an arrhythmogenic milieu capable of facilitating a re-entry circuit. 4 These areas of heterogeneous tissue, more recently termed the "peri-infarct" zone (PIZ) or "gray" zone, are typically located at the transition point between viable myocardium and compact scar and are hypothesized to contain the substrate for slow conduction and fixed/functional block that initiate and maintain VT. 3 Implantable cardioverter-defibrillators (ICDs) can treat re-entrant arrhythmia and have been shown to protect against a high proportion of SCD. 5 Decisions regarding primary prevention ICD insertion currently center around evaluation of New York Heart Association (NYHA) functional class alongside dichotomous assessment of left ventricular (LV) systolic function using an LVEF cutoff of 30% to 35%. 6,7 Typically assessed at a solitary timepoint, this fails to take into account the dynamic nature of LVEF. 1, 8 The everincreasing demand on clinical services represents a parallel issue, and a comprehensive assessment of SCD risk at a singular timepoint (remote from acute myocardial infarction) is desirable. Less dynamic features, which characterize the arrhythmogenic substrate, therefore offer promise.
Late gadolinium enhancement (LGE) cardiac magnetic resonance (CMR) noninvasively identifies dense myocardial fibrosis with high spatial resolution and has good histologic correlation in CAD models. 9 Additionally, quantification of the PIZ by LGE-CMR has been shown to be associated with all-cause mortality [10][11][12][13] and ICD therapy, 14 with more recent studies also evaluating the role of scar microstructure (eg, entropy) in ventricular arrhythmia. 15 There remains, however, a paucity of data describing the application of complex scar analysis to predict SCD in prospectively recruited cohorts with a broad range of LVEF.
Indeed, the utility of noninvasive imaging to predict SCD remains a key research need highlighted by both the American Heart Association and European Society of Cardiology. 6,7 We performed LGE quantification, in combination with bespoke computational analysis of scar microstructure features, to provide a novel mechanistic interrogation into the drivers of SCD in prospectively investigated patients with CAD.

METHODS
STUDY DESIGN. Patients referred to our center for evaluation of ischemic heart disease with LGE-CMR were recruited into a registry between August 2009 and January 2016. These patients were referred from local cardiology clinics in addition to a broad network of specialist and nonspecialist hospitals. The registry complied with the Declaration of Helsinki, and the National Research Ethics Service approved the protocol. All patients provided informed written consent. CMR was undertaken on a 1.5-T scanner (Sonata/ Avanto, Siemens) using a standardized protocol on the day of recruitment or, in a minority of patients, at a prior date during disease work-up. 16 The inclusion criteria for the study were: 1) severe epicardial CAD; 2) prior coronary revascularization; or 3) documented history of prior myocardial infarction (confirmed on CMR Kaplan-Meier curves were plotted to describe the cumulative incidence of the primary outcome by tertiles of PIZ mass and core infarct mass over followup, compared using the log-rank test. To investigate the utility of LGE quantification in the prediction of the primary outcome, we generated univariable and multivariable Cox regression models. The primary multivariable model was adjusted using binary cutoffs of an LVEF of <35% and NYHA functional class of >1 to align with current clinical guidelines for ICD insertion (model A). 7 Competing risk analysis was performed using Fine-Gray subdistribution hazard modeling. Additionally, the incremental predictive value of PIZ mass was examined by calculating categorical (using thresholds of 0%-10%, 10%-20%, and 20%þ to stratify the level of risk) and category-free net reclassification indices.
A secondary Cox regression model was also fitted (model B). To select the covariables in model B, a forward stepwise procedure was applied using a subset of variables in Table 2 with P < 0.10 as the criterion for inclusion, forcing in known predictors of the outcome (age, sex, and LVEF). To prevent the creation of an overly complex model, not all variables that were associated with the primary endpoint on univariable analysis were used in the multivariable model (eg, indexed LV mass and RVEF). In all models, core infarct mass and PIZ mass were subsequently added to assert whether either metric was independently associated with the primary outcome. Model performance was assessed using Harrell's C-statistic.
Forest plots were generated using core infarct mass and PIZ mass per 10 g to aid visual representation.

RESULTS
At baseline, 734 patients were assessed for eligibility, with the final cohort consisting of 437 patients (Supplemental Figure 1). The median interval between CMR and recruitment was 0 days (IQR: 0-0 days), the mean age was 64.4 AE 9.9 years, the mean LVEF was 47.2% AE 16.8%, and 412 (95%) patients had significant CAD or had previously undergone coronary revascularization. Patients were followed up for Transmurality a The extent of spread of LGE emanating outward from the endocardium to epicardium, calculated using a ray tracing method Radiality a Quantification of the circumferential spread of LGE in relation to the center of the LV blood pool.
Number of components a The number of distinct LGE components in a slice (eg, the number of islands of core or PIZ scar) Interface area a The extent (ie, area) of the border zone between myocardium and adjacent LGE Entropy b The level of disorder or heterogeneity within the LGE, calculated by applying standard Shannon entropy 15 The 5 scar microstructure features that were extracted from the LGE images. a Morphologic feature. b Texture feature. LGE a median of 6.3 years (IQR: 5.0-7.9 years). PIZ mass and core mass were strongly correlated (Pearson's correlation coefficient: r ¼ 0.81). Baseline characteristics are described in Table 2 (with extended data in Supplemental Tables 1 and 2). Extended analysis is detailed in the Supplemental Results; reproducibility data for the scar quantification are shown in Supplemental Table 3.
PRIMARY OUTCOME.
Composite of SCD or aborted SCD: Core infarct mass and PIZ mass. At the 10-year follow-up, 49 patients (11.2%) had experienced the primary outcome (20 patients experiencing SCD and 29 patients experiencing aborted SCD). Autopsy results were obtained for 12 of the deaths assigned as SCD. In cases for which autopsy data were not available, SCD was diagnosed by the independent panel of cardiologists using standard endpoint definitions. 19 Cumulative incidence of the primary outcome by tertiles of PIZ mass suggest that patients in higher tertiles had an increased risk of the primary outcome    Table 4.
Composite of SCD or aborted SCD: Scar microstructure analysis. In examining the scar microstructure, each of the 5 features in Table 1 was separately calculated for: 1) core infarct; 2) PIZ; and 3) total scar (core infarct þ PIZ). Figure 3 shows example images highlighting LGE slices with high and low feature values.
All scar microstructure features were associated with the primary endpoint on univariable and multivariable Cox regression analysis, the latter after adjustment for LVEF of <35% and NYHA functional class of >1 (Supplemental Table 5). Figure 4 shows the scar microstructure features with the largest effect estimates. After the addition of core infarct mass and PIZ mass to the Cox regression models, no scar microstructure feature remained significantly associated with the primary endpoint (Supplemental Table 6).

SECONDARY OUTCOMES.
Major heart failure event.    Tables 8 and 9. Reduced LVEF does not predict SCD when LGE quantification data are included in the multivariable models.
Neither PIZ mass nor core infarct mass was associated with major heart failure events on multivariable analysis.
Computational analysis identified a group of clinically plausible scar microstructure features that are associated with SCD; however, these metrics did not independently predict SCD after adjustment for total scar mass.
LGE Zegard et al 24 recently published a retrospective study assessing the association between LGE and SCD in a cohort of CAD patients with a broad range of LVEF. In our study, using a prospectively recruited cohort, we build on the signal that they described in their retrospective registry, demonstrating that both core infarct mass and PIZ mass independently predict SCD. Importantly, however, PIZ mass does not describing a plausible mechanism of action to promote unidirectional conduction block. 18  (A) Core infarct mass and PIZ mass independently predicted SCD after adjustment for parameters used in ICD insertion decisions. (B) PIZ mass and core infarct mass are presented per 10 g. (C) Computational LGE analysis extracted a set of morphologic and texture scar features that were associated with SCD on multivariable analysis (each model adjusted for LVEF of <35% and NYHA functional class of >1). D illustrates the texture-related feature, scar entropy. CMR ¼ cardiac magnetic resonance; ICD ¼ implantable cardioverter-defibrillator; LGE ¼ late gadolinium enhancement; LVEF ¼ left ventricular ejection fraction; NYHA ¼ New York Heart Association; PIZ ¼ peri-infarct zone; SCD ¼ sudden cardiac death.
base of secondary and tertiary hospitals. Our study contains a high percentage of male and White patients, and thus, the results may not be applicable to female patients and non-White populations. We did not assess myocardial ischemia; however, the a priori hypothesis was to evaluate the utility of LGE characterization to predict SCD in stable CAD patients. LGE quantification was only performed using the FWHM method, and recent retrospective studies have highlighted the association between LGE quantification by SD approaches and SCD. 24 The FWHM approach, however, has the highest degree of reproducibility compared to other methods for quantifying LGE, 31 and the majority of PIZ outcome studies have incorporated this methodology. 10 Additionally, we only performed LGE quantification on a single platform, and therefore, the generalizability to other vendors is potentially limited.

CONCLUSIONS
Our study provides novel prospective data demonstrating the value of myocardial fibrosis characterization by CMR to predict SCD in a cohort of stable CAD patients. We also highlight the limitation of LVEF calculation in SCD risk prediction. Multicenter trials should now be considered using appropriate cutoffs for these LGE metrics, above which patients are randomized to ICD insertion or conventional medical therapy.
ACKNOWLEDGMENTS The authors would like to thank our team of medical students who assisted in the follow-up data collection, particularly Won Yoon, Suprateeka Talukder, Aleksandra Lopuszko, and Rohin Reddy. They would also like to acknowledge the excellent team of research nurses in the cardiovascular research center at Royal Brompton Hospital, led by Geraldine Sloane.