Demography and baseline characteristics of all participants

As shown in the flow chart (Fig. 1), a total of 271 patients (age 44.6 ± 14.2 years, 223 (81.9%) men) were enrolled in the final analysis, which included 236 patients without VTA events and 35 patients with VTA events. One hundred and two patients had competing risk events, including HF-related death and heart transplantation. Demographic and baseline characteristics of all patients are summarized in Table 1. There were no statistically significant differences in age, sex, body mass index (BMI), New York Heart Association (NYHA) class, history of hypertension and diabetes between the two groups. Dyspnea (80%) and palpitation (45%) were the most common symptoms. No patients had a history of documented sustained VT, VF, or aborted SCD. LGE was present in 60.5% of (164/271) patients, with the most common pattern of midwall LGE (113/271, 41.7%). The mean LVEF, LVMVR, LGE extent, LVWT, and LVRI of all patients were 22 ± 6.1%, 0.45 ± 0.2, 6.5 ± 8.1%, 8.9 ± 1.8 mm, and 7.7 ± 1.9, respectively. In 80.8% of all patients, the thickest segment of LV myocardial wall was observed in the interventricular septum, including 31.4% in the basal myocardium and 46.8% in the middle myocardium (Fig. 2). Patients with VTA events had significantly higher incidence of left bundle branch block (LBBB) (25.7% vs. 12.2%, p = 0.033), lower LVEDVI (145.3 vs. 163.6 mL/m2, p = 0.028), lower LVMI (60.1 vs. 68.1 g/m2, p = 0.048), thinner LVWT (7.9 vs. 9 mm, p = 0.001), and higher LVRI (8.3 vs. 7.6, p = 0.031) when compared to patients without VTA events. In addition, there were no significant differences in LAEDD, LVEDD, LVESVI, LVMVR, LVEF, and the presence, extent and pattern of LGE between patients with and without VTA events (all p > 0.05). The representative CMR images of two nonischemic DCM patients without and with VTA events are presented in Fig. 3.

Table 1 Clinical characteristics and imaging parameters of all participantsFig. 2

The location distribution of the thickest ventricular wall segments in all patients

Fig. 3

Representative CMR images of two nonischemic DCM patients without (A) and with (B) VTA events. Cine images (A1, A2, B1, B2) and LGE images (A3, A4, B3, B4) on four-chamber long-axis views and basal short-axis slices. Images of 16 myocardial segments reflecting the left ventricular wall thickness (A5, B5). Example cases A 47-year-old male with a lower LVRI calculated as 6.2, who had non-VTA events. LGE images reveal typical septal midwall linear enhancement. B A 19-year-old female with a higher LVRI calculated as 15.2, who had experienced SCD while having no enhancement on LGE images. LVEDV, left ventricular end-diastolic volume; LVWT, left ventricular wall thickness; LVRI, left ventricular remodeling index; CMR, cardiac magnetic resonance; DCM, dilated cardiomyopathy; VTA, ventricular tachyarrhythmia; LGE, late gadolinium enhancement; SCD, sudden cardiac death

Association between CMR parameters and LVRI

The scatterplots indicating the association between CMR parameters and LVRI are shown in Fig. 4. LVRI showed strong correlations with LVEDVI (r = 0.603) and LVESVI (r = 0.617), and weak correlations with LAEDD (r = 0.341), LVEDD (r = 0.418), LVEF (r = −0.418) (all p < 0.0001), and the extent of LGE (r = 0.233, p = 0.003). However, there was no statistically significant association between LVRI and LVMI (r = −0.113, p = 0.063), as well as LVRI and the presence of LGE (r = 0.065, p = 0.287).

Fig. 4

Scatterplots show the correlation between LVRI and LAEDD, LVEDD, LVEF, LVEDVI, LVESVI, LVMI, and the extent of LGE. LVRI, left ventricular remodeling index; LAEDD, left atrial end-diastolic diameter; LVEDD, left ventricular end-diastolic diameter; LVEF, left ventricular ejection fraction; LVEDVI, left ventricular end-diastolic volume index; LVESVI, left ventricular end-systolic volume index; LVMI, left ventricular mass index; LGE, late gadolinium enhancement

LVRI survival analysis in CMR

During a median follow-up of 71 months (interquartile range: 17–134 months), VTA events occurred in 35 patients, including 12 patients with sustained VT, 1 patient with VF, 4 patients with SCD, and 18 patients with aborted SCD.

The optimal cutoff value of LVRI for the primary endpoint was calculated as 7.5 based on the ROC curve analysis (Supplementary Fig. 1), with 71.4% sensitivity, 53.4% specificity and AUC of 0.61 (95% CI: 0.52–0.71). The Kaplan–Meier curves in Fig. 5 showed that patients with LVRI ≥ 7.5 had significantly worse survival from VTA events (log-rank test p < 0.0001). Through 1000 resampling iterations, the bootstrap-derived thresholds showed a median value of 7.7 with an interquartile range of 7.5–7.9, closely clustering around the proposed 7.5 cutoff. Importantly, 75% of the resampled thresholds fell within the clinically reasonable range of 7.5 ± 0.5, suggesting reasonable stability (Supplementary Fig. 2).

Fig. 5

Kaplan–Meier curves for the primary endpoint according to the cutoff value of LVRI (7.5). LVRI, left ventricular remodeling index

The univariable Cox regression analyses demonstrated that LVRI maintained strong associations with VTA events when analyzed both as a dichotomized variable (HR = 4.78, 95% CI: 2.28–10.02, p < 0.001) and as a continuous variable (HR = 1.57 per 1-unit increase, 95% CI: 1.30–1.90, p < 0.001) in models that did not account for competing risks. The results of univariable and multivariable competing risk regression analyses for VTA events are presented in Table 2. In the univariable competing risk regression analyses, LVRI showed significant associations with VTA events both as a dichotomized variable (≥ 7.5: HR 2.82, 95% CI: 1.37–5.79, p = 0.005) and as a continuous variable (per 1-unit increase: HR 1.21, 95% CI: 1.02–1.43, p = 0.026), which align well with the univariate Cox regression analyses. In contrast, LVMVR, the presence, extent and pattern of LGE were not associated with VTA events (all p > 0.05). In the multivariable competing risk regression model 1, when heart transplantation and HF-related death were counted as competing risks, LVRI ≥ 7.5 (adjusted HR 2.496 [95% CI: 1.213–5.138], p = 0.013) was an independent predictor of VTA events. The bootstrap procedure confirmed minimal shrinkage (ΔHR = +1.402%) with a consistent effect size (original HR = 2.496 vs. bootstrapped HR = 2.529, 95% CI: 1.24–6.20) in the primary predictor LVRI. Other clinically relevant variables like LVMI maintained proximate value despite modest shrinkage (7.839%, original HR = 0.983 vs. bootstrapped HR = 0.982, 95% CI: 0.97–1). The bootstrap results (Supplementary Table 1 and Supplementary Fig. 3) support the structural robustness of the original competing risk regression model. However, when LVRI was included as a continuous variable in model 2, the results showed that LVMI was the only independent predictor of VTA events (adjusted HR 0.983 [95% CI: 0.968–0.998], p = 0.031), after adjusting for age, sex, LBBB, and LVMI.

Table 2 Uni- and multivariable competing risk regression analyses for VTA events with clinical and CMR parametersIncremental prognostic values of LVRI over conventional CMR parameters

Different Cox models were utilized to further investigate the prognostic significance of LVRI in predicting VTA among patients with nonischemic DCM with LVEF < 35%. The chi-square values for Model 1 (LVEF), Model 2 (LVEF + LGE presence), and Model 3 (LVEF + LGE presence + LVRI) were 0.0004, 2.1 (p = 0.149 vs. Model 1), and 27.1 (p < 0.001 vs. Model 2), respectively. The incorporation of LVRI provided significant incremental predictive discrimination (C-index: 0.75 [0.67–0.84]) with clinically meaningful reclassification (IDI = 0.144, p < 0.001) (Fig. 6). The analyses of Model 4 (LVEF + LGE extent) and 5 (LVEF + LGE extent + LVRI) were repeated using LGE extent instead of LGE presence. LVRI demonstrated similar incremental predictive value (Chi-square improvement from 3.06 to 20.55, p < 0.001; C-index 0.75, 95% CI: 0.67–0.83; IDI = 0.14; all p < 0.001), as shown in Supplementary Fig. 4.

Fig. 6

Incremental prognostic value of LVRI over conventional LVEF and LGE presence to VTA endpoints. LVRI, left ventricular remodeling index; LVEF, left ventricular ejection fraction; LGE, late gadolinium enhancement; VTA, ventricular tachyarrhythmia; IDI, integrated discrimination improvement

Reproducibility analysis

Quantification of LVWT had good reproducibility in DCM patients, as the interobserver ICC (0.981 [95% CI: 0.953–0.993]) was higher than 0.75.