Abstract 4144385: Advanced Cardiac MRI Parameters Phenomapping to Predict Outcomes in Patients with Nonischemic Cardiomyopathy

Background: Non-ischemic cardiomyopathy is a classification that includes several heterogenous phenotypes. Derivation of distinct phenotypes may enable improved risk stratification. Comprehensive functional, morphologic, and substrate assessment of the myocardium can be leveraged to derive distinct phenogroups through unsupervised latent class/cluster analysis but requires accurate and reproducible imaging biomarkers. We sought to evaluate the impact of cardiac magnetic resonance (CMR) imaging parameters to derive distinct phenotypes in NICM patients. Methods: Consecutive NICM patients referred to CMR between 2001 and 2017 were evaluated. Latent Class Analysis (LCA) was performed to identify phenotypes enriched by comprehensive CMR assessment. The primary endpoint was death, heart transplant, or left ventricular assist device implantation. Kaplan–Meier survival curves were estimated, and differences between phenotypic groups were assessed by the log-rank test. Decision tree modeling was used to predict the latent class membership. LCA and Decision tree analysis were completely blinded to the clinical outcomes. Results: A total of 458 patients were identified (53.2 ± 15.7 years, 40.7% females), with 83 primary events. LCA identified 3 distinct clusters of patients, with significant differences in overall outcome (p<0.0001, Figure 1A). Principal component analysis demonstrated 3 well demarcated clusters. Weighted variable analysis demonstrates the variable importance for determining phenogroup status (Figure 1B). Decision tree analysis (Figure 1C) demonstrated high accuracy (0.89) to predict LCA phenogroup status. Conclusion: CMR enriched phenomapping identified 3 distinct NICM phenogroups associated with significant differential risk profiles, with advanced CMR measures including comprehensive myocardial strain assessment being important imaging parameters used for phenogroup stratification. Decision tree analysis provided an accurate, simplified, and explainable method to enable a novel approach for patient selection. Future studies are needed to determine if distinct phenogroups are associated with differential treatment response to various interventions.