Background: Patients undergoing valvular heart surgery show substantial clinical heterogeneity that may not be fully captured by conventional measures such as left ventricular ejection fraction (LVEF). Unsupervised clustering may identify clinically meaningful phenotypes. Objectives: To identify data-driven clusters of the early postoperative rehabilitation course by integrating clinical, echocardiographic, and functional data, collected before and during inpatient rehabilitation, and to compare LVEF and early rehabilitation outcomes across phenotypes. Methods: Patients admitted after valvular heart surgery were retrospectively analyzed. Principal Component Analysis followed by K-means clustering was applied to demographic, cardiovascular, valvular, echocardiographic, pharmacological, and functional variables obtained from medical history, admission assessments and in-hospital complications. Cluster selection was based on the silhouette score. Between-cluster comparisons were performed using non-parametric and categorical statistical tests with post-hoc correction. Results:The cohort included 298 patients, of whom 59.7% were male, with a median age of 73 years [IQR=13]. Three clinically interpretable clusters were identified. C1 (n = 72, 24.2%) showed predominant aortic insufficiency, preserved autonomy, and low atrial arrhythmic burden. C2 (n = 117, 39.3%) was dominated by aortic stenosis, with greater cardiometabolic burden and greater wall-thickness measures. C3 (n = 109, 36.6%) was dominated by mitral insufficiency, with more frequent mitral valve prolapse, higher atrial arrhythmic burden, and an altered mitral inflow profile. LVEF did not differ significantly across phenotypes. In complete-case analyses, Six-Minute Walk Test improvement showed a preliminary between-phenotype difference, mainly driven by greater gains in C1 than C2. Conclusions: Unsupervised clustering identified three valve-substrate-associated clinical profiles. Our findings suggest that multidimensional profiling may provide an integrated characterization of postoperative heterogeneity and early rehabilitation trajectories after valvular heart surgery. Given the retrospective design, modest sample size, and complete-case outcome analysis, these hypothesis-generating findings require validation in larger prospective cohorts.
Multidimensional profiling of the early postoperative rehabilitation course after valvular heart surgery using unsupervised clustering: an exploratory analysis / Marra, E., Liuzzi, P., Mannini, A., Falorsi, G., Buccione, E., Nicolosi, B., Amato, S., Magi, C.E., Iovino, P., Vellone, E., Braga, J., Macchi, C., Dollaku, H.. - In: FRONTIERS IN MEDICINE. - ISSN 2296-858X. - ELETTRONICO. - 13:(2026), pp. 0-0. [10.3389/fmed.2026.1929975]
Multidimensional profiling of the early postoperative rehabilitation course after valvular heart surgery using unsupervised clustering: an exploratory analysis
Falorsi, Gaia;Buccione, Emanuele;Nicolosi, Biagio;Magi, Camilla Elena;Iovino, Paolo;Macchi, Claudio;Dollaku, Hamilton
2026
Abstract
Background: Patients undergoing valvular heart surgery show substantial clinical heterogeneity that may not be fully captured by conventional measures such as left ventricular ejection fraction (LVEF). Unsupervised clustering may identify clinically meaningful phenotypes. Objectives: To identify data-driven clusters of the early postoperative rehabilitation course by integrating clinical, echocardiographic, and functional data, collected before and during inpatient rehabilitation, and to compare LVEF and early rehabilitation outcomes across phenotypes. Methods: Patients admitted after valvular heart surgery were retrospectively analyzed. Principal Component Analysis followed by K-means clustering was applied to demographic, cardiovascular, valvular, echocardiographic, pharmacological, and functional variables obtained from medical history, admission assessments and in-hospital complications. Cluster selection was based on the silhouette score. Between-cluster comparisons were performed using non-parametric and categorical statistical tests with post-hoc correction. Results:The cohort included 298 patients, of whom 59.7% were male, with a median age of 73 years [IQR=13]. Three clinically interpretable clusters were identified. C1 (n = 72, 24.2%) showed predominant aortic insufficiency, preserved autonomy, and low atrial arrhythmic burden. C2 (n = 117, 39.3%) was dominated by aortic stenosis, with greater cardiometabolic burden and greater wall-thickness measures. C3 (n = 109, 36.6%) was dominated by mitral insufficiency, with more frequent mitral valve prolapse, higher atrial arrhythmic burden, and an altered mitral inflow profile. LVEF did not differ significantly across phenotypes. In complete-case analyses, Six-Minute Walk Test improvement showed a preliminary between-phenotype difference, mainly driven by greater gains in C1 than C2. Conclusions: Unsupervised clustering identified three valve-substrate-associated clinical profiles. Our findings suggest that multidimensional profiling may provide an integrated characterization of postoperative heterogeneity and early rehabilitation trajectories after valvular heart surgery. Given the retrospective design, modest sample size, and complete-case outcome analysis, these hypothesis-generating findings require validation in larger prospective cohorts.I documenti in FLORE sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.



