Postdoctoral Fellow University of Pittsburgh Pittsburgh, Pennsylvania, United States
Background: Pediatric hypertrophic cardiomyopathy (HCM) presents unique challenges due to its heterogeneous clinical manifestations and outcomes. This study aimed to leverage advanced echocardiographic techniques and unsupervised machine learning (ML) to identify distinct phenogroups based on hemodynamic parameters, with implications for personalized risk stratification and management.
METHODS AND RESULTS: A retrospective, single-center analysis of 60 pediatric HCM patients (46.7% male, median age 12.3 [7.8–14.5] years) was conducted. Diagnosis adhered to 2020 AHA/ACC guidelines, with patients classified into obstructive and non-obstructive HCM. Speckle-tracking echocardiography (STE) was used to assess segmental strain, myocardial work (MW), and pressure-strain loops. Global work indices, including global constructive work (GCW), global wasted work (GWW), and global work efficiency (GWE), were also derived. Clinical outcomes, including sudden cardiac death (SCD), implantable cardioverter defibrillator (ICD) discharges, and ventricular tachycardia (VT), were tracked over a median follow-up of 4.9 years. Principal component analysis (PCA) and K-means clustering were employed to stratify patients into phenogroups based on their hemodynamic profiles. Three phenogroups were identified through unsupervised ML: Cluster 1 (28.3%, n=17), Cluster 2 (38.3%, n=23), and Cluster 3 (33.3%, n=20). Significant demographic differences were found in age and sex between clusters (Table). In terms of echocardiographic parameters, Cluster 1 demonstrated significantly impaired global longitudinal strain (GLS: -14.3±2.1 vs. -21.0±2.5 and -21.7±2.6 in Clusters 2 and 3, p< 0.001). These differences were apparent in the basal-, mid- and apical strain patterns throughout the cardiac cycle (Figure). Furthermore, Cluster 1 had higher mechanical dispersion (MD: 68.7 [52.5–112] ms vs. 42.3 [33.8–57.8] ms and 40.6 [36.0–59.9] ms, p=0.001). Pressure-strain loop analysis revealed a triangular pattern in Cluster 3, consistent with shortened isovolumic contraction and relaxation phases. Then, during follow-up, differences in outcomes between clusters became apparent. Cluster 1 had a significantly higher cumulative incidence of adverse outcomes (19.6%, 95% CI: 0.0–37.3%) compared to Cluster 2 (4.3%) and Cluster 3 (0%) (log-rank p=0.013). Outcomes in Cluster 1 included ICD discharges (n=6), 3 for sustained VT and 3 for SCD, and two deaths, both attributed to cardiac causes.
Conclusion: Unsupervised ML, based on echocardiographic hemodynamic parameters, successfully identified three phenogroups in pediatric HCM. Cluster 1, characterized by impaired myocardial strain, increased mechanical dispersion, and inefficient myocardial work, had the highest incidence of adverse events, suggesting that distinct hemodynamic profiles may serve as a basis for personalized risk assessment in pediatric HCM. This approach offers a promising avenue for refining management strategies and improving outcomes.
