Rapid Fire Abstracts
Carl Ammoury, MD
Research Fellow
Cleveland Clinic Foundation
Carl Ammoury, MD
Research Fellow
Cleveland Clinic Foundation
Yanjun Wu, MSc
Research Data Scientist
Lerner Research Institute
Theerawat Korkerdsup, MD
Advance imaging cardiologist
Bangpakok 9 International Hospital , Thailand
Tom Kai Ming Wang, MBChB MD
Cardiologist
Cleveland Clinic
Diane Rizkallah, MD
Research Fellow
Cleveland Clinic Foundation
Kashyap Bodi, MD
Research Fellow
Cleveland Clinic, Cleveland, OH, United States
Tess Calcagno, MD
Resident
Cleveland clinic
David Chen, PhD
Director of Artificial Intelligence
Cleveland Clinic
Christopher Nguyen, PhD, FSCMR, FACC
Director, Cardiovascular Innovation Research Center
Cleveland Clinic
Wilson Tang, MD
Cardiologist
Cleveland Clinic
Xiaofeng Wang, PhD
Full Staff Member/Data Scientist/Professor
Cleveland Clinic Foundation
Deborah Kwon, MD, FSCMR
Director of Cardiac MRI
Cleveland Clinic
Cluster analysis leverages computational capabilities of machine learning to identify patients with distinct phenotypes and enhance risk stratification. We sought to highlight the impact of comprehensive cardiac magnetic resonance (CMR) imaging feature tracking assessment on conventional measurements to enrich phenomapping analysis in patients with nonischemic cardiomyopathy (NICM).
Methods:
NICM patients referred to CMR between 2001 and 2017 were evaluated. Phenomapping analysis was performed using Latent Class Analysis (LCA). Phenogroups were identified using conventional CMR parameters, with a subgroup analysis in patients with additional advanced parameters (comprehensive feature tracking (FT) assessment and left atrioventricular coupling index (LACI)). The primary outcome was death, heart transplant, or LVAD implantation. Kaplan-Meier analysis was performed, and log-rank test statistics were compared to assess the incremental effect of advanced CMR parameters. Decision tree and weighted analyses were performed to identify the parameters most responsible to derive the separate clusters along with their respective thresholds. Stability of phenotype identification was evaluated using confusion matrix analysis.
Results: Three levels of cluster analysis were performed: 1) cluster 1: 633 patients (52.6 ± 16.0 years, 40.4% females) with conventional measurements, 2) 458 patients with conventional 3) 458 patients with comprehensive FT and LACI. In cluster 1, conventional CMR measurements in the larger cohort did not show clear cluster distinction, but phenogroup status was associated with significantly differing survival. (Figure 1A). Subgroup analysis demonstrated stable phenogroup derivation with the same conventional CMR measures in a smaller cohort with available comprehensive FT evaluation. Overall accuracy of the subgroup LCA derivation in cluster 2 was very high: 96.9% with a Kappa statistic of 0.85 (Figure 1B). Cluster analysis 3 resulted in significantly distinct phenogroups that were associated with significantly more distinct risk profiles (40.85, p-value< 0.0001) compared to conventional CMR parameters (8.98, p-value= 0.011). Weighted analysis identified left atrial (LA) reservoir strain, LV Ejection Fraction, and Late Gadolinium enhancement (LGE) as the strongest predictors for clustering (Figure 2). Decision tree identified a threshold of 28% for LV Ejection Fraction and 14% for LA reservoir strain with an accuracy of 89% (Figure 3), compared to the unsupervised LCA.
Conclusion:
Comprehensive FT enables more robust phenotype derivation, resulting in significantly improved risk stratification in patients with non-ischemic cardiomyopathy. Future studies are needed to evaluate the distinct underlying mechanical signatures and differential response to therapeutic interventions.
