Anticoagulation management in intracerebral hemorrhage patients with deep vein thrombosis: insights from unsupervised machine learning and nomogram analysis

Introduction Lower extremity deep vein thrombosis (DVT) is a frequent complication in patients with intracerebral hemorrhage (ICH), increasing the risk of adverse outcomes and mortality. However, standard anticoagulation therapy can lead to hematoma expansion, highlighting the need for reliable and practical risk assessment tools. While unsupervised machine learning has shown promise in patient stratification, its clinical applicability is limited. This study integrates unsupervised machine learning with nomogram analysis to identify risk factors and establish a clinically actionable risk assessment tool. Methods The study was conducted in two phases. In the retrospective exploratory phase, 191 ICH patients receiving anticoagulation were grouped using K-means and hierarchical clustering. Incidence rates of DVT and adverse events were analyzed to identify key risk factors influencing anticoagulant safety. A nomogram was then constructed to quantify adverse event risk. In the prospective validation phase, 291 patients were stratified into high- and low-risk groups based on nomogram scores. VTE and adverse event rates were compared between groups, with multivariate regression and subgroup analyses performed. Results Key risk factors identified included admission mRS, GCS, and ADL scores, admission and discharge ICH volume, and admission albumin level. In the validation cohort, the low-risk group had significantly lower VTE (16.9% vs. 65.1%, p 0.001) and adverse event rates (13.4% vs. 46.3%, p 0.001) than the high-risk group. Multivariate regression confirmed a significant inverse association between low-risk classification and occurrence of VTE and adverse events. Conclusion This study demonstrates that unsupervised machine learning, combined with a nomogram, can effectively stratify risk in ICH patients receiving anticoagulation. The risk assessment tool reliably identifies patients at lower risk of adverse outcomes, supporting safer and more individualized clinical decision-making.