Cardiac magnetic resonance-based layer-specific strain in immune checkpoint inhibitor-associated myocarditis

Abstract

Aims: To assess the different imaging characteristics between corticosteroid-sensitive (CS) and corticosteroid-refractory (CR) immune checkpoint inhibitor-associated myocarditis (ICIaM) with cardiac magnetic resonance (CMR) and the potential CMR parameters in the early detection of CR ICIaM.

Methods and results: Thirty-five patients diagnosed with ICIaM and 30 age and gender-matched cancer patients without a history of ICI treatment were enrolled. CMR with contrast was performed within 2 days of clinical suspicion. Left ventricular ejection fraction (LVEF) and late gadolinium enhancement (LGE) were assessed by CMR. LV sub-endocardial (GLSendo) and sub-epicardial (GLSepi) global longitudinal strains were quantified by offline feature tracking analysis. CS and CR ICIaM were defined based on the trend of Troponin I and clinical course during corticosteroid treatment. All 35 patients presented with non-fulminant symptoms upon initial assessment. Twenty patients (57.14%) were sensitive, and 15 (42.86%) were refractory to corticosteroids. Compared with controls, 22 patients (62.86%) with ICIaM developed LGE. LVEF decreased in CR ICIaM compared with the CS group and controls. GLSendo (-14.61 ± 2.67 vs. -18.50 ± 2.53, P < 0.001) and GLSepi (-14.75 ± 2.53 vs. -16.68 ± 2.05, P < 0.001) significantly increased in patients with CR ICIaM compared with the CS ICIaM. In patients with CS ICIaM, although GLSepi (-16.68 ± 2.05 vs. -19.31 ± 1.80, P < 0.001) was impaired compared with the controls, GLSendo was preserved. There was no difference in CMR parameters between LGE-positive and negative groups. LVEF, GLSendo, and GLSepi were predictors of CR ICIaM. When LVEF, GLSendo, and GLSepi were included in multivariate analysis, only GLSendo remained an independent predictor of CR ICIaM (OR: 2.170, 95% CI: 1.189-3.962, P = 0.012). A GLSendo of ≥-17.10% (sensitivity, 86.7%; specificity, 80.0%; AUC, 0.860; P < 0.001) could predict CR ICIaM in the ICIaM cohort. Kaplan-Meier analysis showed that in patients with impaired GLSendo of ≥-17.10%, cardiovascular adverse events (CAEs) occurred much earlier than in patients with preserved GLSendo of <-17.10% (Log-rank test P = 0.017).

Conclusions: CR and CS ICIaM demonstrated different functional and morphological characteristics in different myocardial layers. An impaired GLSendo could be a helpful parameter in early identifying corticosteroid-refractory individuals in the ICIaM population.

Keywords: Cardiac magnetic resonance; Global longitudinal strain; Immune checkpoint inhibitor; Late gadolinium enhancement; Myocarditis.

Figure 1

The presence of LGE in ICIaM. In one patient with CR ICIaM (A), LGE was located in the sub‐epicardial layer of the LV anterior and anterolateral wall. In one patient with CS ICIaM (B), LGE was located in the sub‐epicardial layer of the LV anteroseptal and inferior wall. Yellow arrows: LGE location.

Figure 2

Offline tissue tracking analysis of strain parameters. (A) A region of interest covering the LV myocardium was manually drawn by delineating LV's endo‐ and epicardial borders at the end‐diastolic phase of all cine images. Red line: Sub‐endocardial borders; Green line: Sub‐epicardial borders; Blue line: a marker of the LV long axis. (B) GLS‐time curves for GLSendo (red curve) and GLSepi (green curve). The white vertical line reflects the peak strain value and its corresponding time.

Figure 3

Two examples of cTnI trajectory in CR (A) and CS (B) ICIaM patients during corticosteroid‐based treatment. (A) CR ICIaM with up‐trending cTnI despite high‐dose methylprednisolone. (B) CS ICIaM with down‐trending cTnI. Blue solid line with yellow dots: daily cTnI level; green dashed line: trend line; red solid line: timing of CMR and treatment adjustment.

Figure 4

Kaplan–Meier curves for cardiovascular adverse events by GLSendo (A) and LGE (B).

Figure 5

(A) ROC curve analysis of GLSendo, GLSepi, and LVEF in differentiating CR from CS ICIaM. LVEF could not robustly distinguish CR ICIaM from CS ICIaM (AUC, 0.677; P = 0.059). The AUC of GLSendo was significantly larger than the AUC of GLSepi (0.860 vs. 0.717, P = 0.035). (B) ROC curve analysis of GLSepi in differentiating CS ICIaM from controls. The point corresponding to the cutoff value was marked as a yellow circle on each curve.