Phenotypic clustering using cardiac magnetic resonance imaging differentiates genetic background and outcomes in dilated cardiomyopathy

Background

Dilated cardiomyopathy (DCM) is a heterogeneous condition with diverse imaging phenotypes, genetic substrates and clinical outcomes. Identifying phenotypic subgroups using cardiac magnetic resonance (CMR) may improve risk stratification and guide therapy.

Purpose

To identify distinct CMR-based phenotypic clusters in a well-characterized DCM cohort and evaluate their association with genetic background and clinical outcomes.

Methods

This single-centre retrospective study included DCM patients who underwent CMR imaging and genetic testing. Medical records were reviewed for genetic and clinical outcomes. The primary endpoint was a composite of sustained ventricular tachycardia, appropriate implantable cardioverter-defibrillator therapy, or sudden cardiac death. The secondary endpoint was a composite of heart transplantation, left ventricular assist device implantation or pump failure death. An unsupervised k-means clustering algorithm was applied to standardized (z-scored) CMR variables, including indexed LV end-diastolic volume (LVEDVi), indexed LV mass (LVMassi), LV ejection fraction (LVEF), RV/LV end-diastolic volume ratio, RV ejection fraction (RVEF), indexed left atrial area (LAAi), native T1 time and number of segments with late gadolinium enhancement (LGEn) Event-free survival was assessed using Kaplan–Meier analysis with log-rank testing, and prevalence of likely pathogenic or pathogenic high-risk arrhythmic variants (according to ESC guidelines) was compared using Pearson’s chi-square test.

Results

Among 271 patients (mean age 51 years ±14 [SD], 70% male) 3 distinct phenotypic clusters were identified. Cluster 1 (n=153) demonstrated higher LVEF, smaller LV volumes, and minimal fibrosis. Cluster 2 (n=72) showed larger LV dimensions and mass, lower LVEF and intermediate fibrosis. Cluster 3 (n=46) had smaller LV volumes but markedly higher focal and interstitial fibrosis.

During a median follow-up of 34 months, the primary endpoint occurred in 17 (6.3%) and the secondary in 14 (5.1%) patients. For both endpoints, Cluster 1 showed significantly better event-free survival compared with Cluster 2 (p = 0.019 for the primary and p < 0.001 for the secondary endpoint) and Cluster 3 (p = 0.004 and 0.015, respectively). No significant difference was observed between Clusters 2 and 3 (p = 0.292 and 0.185), although the Kaplan–Meier curves suggested a divergent trend for the primary endpoint, with the highest arrhythmic burden observed in Cluster 3. The prevalence of high-risk arrhythmic variants differed significantly across clusters (χ²(2) = 6.425, p = 0.040), mainly driven by a higher proportion in Cluster 3 (19.6%) compared with Clusters 1 (7.2%) and 2 (8.3%).

Conclusion

Unsupervised CMR-based clustering identifies distinct DCM phenotypes that align with differences in genetic background and clinical outcomes. These findings validate the use of CMR phenotyping as a tool for data-driven risk stratification in DCM.Phenotypic ClustersKaplan-Meier curves for both endpoints