Distinct clinical phenotypes in acute heart failure identified by unsupervised clustering: a nationwide registry-based study.

Background: Acute heart failure (AHF) is a highly heterogeneous clinical syndrome, posing challenges for risk stratification and management. Phenotype-based classification using unsupervised clustering may provide insights beyond conventional approaches. Methods: We analyzed a nationwide, prospective, multicenter registry of patients hospitalized with AHF. Baseline demographic characteristics, clinical variables, laboratory findings, and echocardiographic parameters obtained at index hospitalization were used for unsupervised clustering. K-means clustering was applied, and the optimal number of clusters was determined based on internal validation metrics and clinical interpretability. Clinical characteristics, treatment patterns, and outcomes were compared across clusters. Cox proportional hazards models were used to evaluate the association between cluster membership and clinical outcomes. Results: A total of 7,351 patients were classified into five distinct clinical phenotypes. The clusters demonstrated marked differences in baseline characteristics, comorbidity burden, hemodynamic profiles, and cardiac function. Treatment patterns, including guideline-directed medical therapy and supportive interventions, varied significantly across clusters. During follow-up, all-cause mortality differed significantly among clusters ( Conclusion: In this large, nationwide cohort of patients with AHF, unsupervised clustering identified five clinically distinct phenotypes with significantly different characteristics, management patterns, and outcomes. These findings support the clinical relevance of phenotype-based classification and highlight its potential to enhance risk stratification and inform more personalized management strategies in AHF.