Distinct phenotypic clusters identified by unsupervised network profiling in secondary mitral regurgitation
Y Kang, K Y Liu, Y J Xie, R X Zhang, G Y Yuo, X J Chen, Q ZhangAbstract
Background
Secondary mitral regurgitation (SMR) typically results from left ventricular dilation with or without systolic dysfunction irrespective of underlying etiologies. The morphological and functional components and underlying diseases which may account for the clinical impact of SMR have not been systematically assessed.
Purpose
The aim of this work was to identify distinct phenotypes by unsupervised network profiling the key echocardiographic and clinical features in SMR.
Methods
Clinical and echocardiographic features network profiling were performed on a cohort of hospitalized patients with SMR if they had left ventricular dilation with or without decreased ejection fraction (LVEF). Clinical features include age, sex, body mass index and diagnosis captured by ICD-10. Laboratory and echocardiographic features were included. Principal component (PC) analysis and subsequent UMAP analysis were used to condense these into mogeneous clusters. Clusters were tested for their clinical features.
Results
A total of 1586 patients were enrolled, with a median age of 63±15 years and 63% men. After standardization and dimensionality reduction (with PCA retaining 95% variance, together with t-SNE and UMAP), nine clustering algorithms were applied with systematic parameter tuning. After excluding highly unbalanced solutions (minimum cluster proportion <5%), 65 candidate models remained. Performance was assessed using silhouette score, Calinski-Harabasz index, and Davies-Bouldin index. The optimal model was UMAP combined with K-means (k=4), with a silhouette score of 0.503, CH index of 2024.671, and DB index of 0.648. Four distinct phenotypic clusters were created (Figure 1 left) and their echocardiographic features are illustrated in Figure 1 (right). The integrated clinical phenotypes are summarized in Figure 2. On summary, cluster 1 represented the dilated cardiomyopathy (DCM) phenotype, characterized by the youngest age, severe LV dilatation, lowest LVEF, and highest prevalence of DCM. Cluster 2 represented elderly cardiopulmonary phenotype, with the oldest age, extremely high chronic obstructive pulmonary disease prevalence, and high pulmonary hypertension rate. Cluster 3, ischemic cardiomyopathy phenotype, with near-universal coronary artery disease, highest cardiovascular event rate, and highest troponin-T level. Cluster 4, hypertension-cardiorenal phenotype, with female predominance, preserved LVEF, and highest prevalence of hypertension, chronic kidney disease, and elevated natriuretic peptides.
Conclusions
Unsupervised clustering identified four distinct phenotypes in patients with SMR, i.e., a young DCM phenotype, an elderly cardiopulmonary phenotype, an ischemic cardiomyopathy phenotype, and a hypertension and cardiorenal phenotype. Clustering effectively differentiates ischemic from non-ischemic cardiomyopathy and identifies two novel underrecognized phenotypes that may enable more personalized therapeutic strategies.SMR clusteringFeature profiles of the four clusters