Risk prediction of traumatic and non-traumatic subdural hematoma after ischemic stroke: a nationwide competing-risk model with external validation.

Subdural hematoma (SDH) represents a growing hemorrhagic concern among ischemic stroke survivors receiving long-term antithrombotic therapy. Although aging, multimorbidity, and antithrombotic exposure are all contributors to SDH, there is limited evidence on subtype-specific risk factors in stroke populations, and existing studies have often overlook competing mortality. Practical tools to guide individualized antithrombotic stewardship remain absent from contemporary guidelines. A nationwide retrospective cohort study was conducted using Korean National Health Insurance Service data from 2010 to 2023. Adults with incident ischemic stroke were followed for traumatic and non-traumatic SDH using Fine-Gray competing risk regression and cause-specific hazard models. Variables were selected using the Akaike information criterion to develop integer-based prediction scores. Internal validation used cross-validation, and external validation applied the models to an independent ischemic stroke cohort from the UK Biobank. A web-based calculator was employed to facilitate clinical use. Among 506,746 ischemic stroke survivors (median follow-up 57.9 months), 7,425 developed traumatic SDH and 1,705 developed non-traumatic SDH. The 10-year cumulative incidences were 1.94% for traumatic and 0.43% for non-traumatic SDH. Older age, multimorbidity, and antithrombotic exposure were independently associated with increased risk. Traumatic SDH demonstrated stronger associations with antiplatelet therapy and diabetes, whereas non-traumatic SDH was most strongly linked to warfarin use. Warfarin emerged as the only modifiable risk factor. Prediction models showed clear risk-gradient separation for both subtypes and maintained stratification performance in the UK Biobank cohort. A web-enabled platform provided individualized 10-year cumulative incidence estimates using routinely available clinical variables. This nationwide competing-risk analysis identified subtype-specific associations of aging, comorbidities, and antithrombotic exposure with SDH risk among ischemic stroke survivors. Because the models predict SDH specifically rather than overall intracranial hemorrhage, they are intended to support risk-stratified surveillance, monitoring, and patient counseling rather than to guide antithrombotic cessation in isolation, as such decisions require balancing ischemic recurrence against the full spectrum of major bleeding risk.