Gower–PAM-Based Phenotyping in Ischaemic Stroke: Clinical, Cardiac, and Biochemical Integration and Relationships with Vertebrobasilar Patterns

Ischaemic stroke is phenotypically heterogeneous, and cardiac and metabolic characteristics may influence its clinical course. This study integrated clinical, echocardiographic, and biochemical data to derive data-driven phenotypes and examine their relationship with vertebrobasilar involvement patterns. In 150 patients with posterior-circulation ischaemic stroke, binary and continuous clinical, cardiac, and biochemical variables were integrated using Gower's distance to construct a scale-independent distance matrix. Unsupervised clustering was performed with the Partitioning Around Medoids (PAM) algorithm, and the optimal solution was selected according to the highest average silhouette width. Inter-cluster differences were assessed using prevalence estimates and odds ratios (95% CI) for binary variables and standardised mean differences for continuous variables, with Benjamini Hochberg false discovery rate (FDR) adjustment. Cluster separation was visualised by classical multidimensional scaling. Two phenotypes were identified. Cluster 2 (n=49) showed a cardiometabolic/atrial valvular profile, characterised by higher rates of diabetes mellitus, coronary artery disease, hyperlipidaemia, atrial fibrillation, and valve replacement/warfarin use, with a more atherogenic glycaemic biochemical profile: higher total cholesterol, LDL cholesterol, triglycerides, Lp(a), and glucose, and lower HDL cholesterol. Left atrial dilatation was more frequent in Cluster 2 (75.5% vs 35.0%; p<0.001, q<0.001). Cluster 1 (n=101) demonstrated a diastolic phenotype, with more frequent left ventricular diastolic dysfunction (65.3% vs 18.4%; OR 8.38, 95% CI 3.65–19.2; p<0.001, q<0.001). Among vertebrobasilar patterns, only right vertebral artery distal-segment stenosis remained significant after correction (21.8% vs 2.0%; OR 13.4, 95% CI 1.75–102; p=0.003, q=0.041). Gower–PAM phenotyping identified two meaningful phenotypes with shared cerebrovascular topography but divergent substrates, supporting cluster-specific secondary prevention.