Data-driven phenotyping of ischemic stroke using cluster analysis: association with long-term stroke recurrence

Effective secondary prevention for ischemic stroke (IS) requires accurate risk stratification and individualized treatment strategies. This study aimed to identify distinct IS phenotypes using a systematic, data-driven clustering approach based on clinical, demographical, and imaging features, and compared to recurrent stroke across clusters using survival analysis. Patients hospitalized with diagnosis of IS in the stroke unit of Istanbul Faculty of Medicine between 2011 and 2018 were screened, and those who attended three-month follow-up outpatient clinic were included. Clustering analysis were performed using Gower distance on mixed dataset followed by partitioning around medoids for clustering. Variable selection was conducted using a systemic approach combining univariate methods, internal validity indices, and visualization techniques. 799 patients (60% male) with mean age 63.53 years (± 15.17) were included in the study. The dataset was ultimately classified into four clusters. Cluster-1 represented younger male patients with favorable in-hospital course, Cluster-2 captured high-risk male patients with substantial vascular comorbidity and inflammation, exhibiting the highest recurrence; Cluster-3 comprised older female patients with cancer and inflammatory profiles, and Cluster-4 composed of the oldest female patients with atrial fibrillation and prior stroke, characterized by more severe stroke. The median follow-up duration was 26 months, and recurrent stroke differed significantly among the four clusters. These findings suggest that cluster analysis may serve as a valuable tool for clinical risk stratification in IS. The higher frequency of recurrent stroke in Cluster 2, characterized by greater vascular comorbidities indicates that this Cluster may represent a population requiring closer monitoring and more intense secondary prevention strategies.