Identifying why a stroke occurred is one of the most consequential — and elusive — clinical questions in cerebrovascular medicine. The answer shapes antithrombotic therapy, secondary prevention, and long-term prognosis. For the roughly one-in-four ischemic stroke patients classified as embolic stroke of undetermined source (ESUS), that answer remains frustratingly out of reach, and outcomes suffer as a result.
Drawing on the INSIGHT registry (NCT04693767), investigators analyzed 388 patients across cardioembolic (CE), atrial fibrillation-specific (AFIB), and non-CE groups, examining 123 features spanning demographics, laboratory values, medication history, and physical thrombus characteristics retrieved during thrombectomy. Using correlations, univariate analyses, and LASSO-regularized machine learning models, the team identified several variables that independently tracked with stroke etiology. Elevated baseline prothrombin time and INR — specifically in patients not already anticoagulated — along with baseline use of antiarrhythmics and antihypertensives, and higher pre-stroke modified Rankin Scale scores, were each associated with elevated probability of cardioembolic and AFIB-driven events. Conversely, higher baseline platelet and red blood cell counts correlated with reduced odds of AFIB etiology, suggesting a distinct hematological profile for large-artery or cryptogenic mechanisms. LASSO models demonstrated meaningful predictive accuracy for both CE and AFIB classification.
This work matters because thrombus composition — the physical signature of a clot — has long been theorized to encode etiological information, yet clinical adoption of such biomarkers has been slow. The INSIGHT dataset is relatively modest at 388 patients, and the registry's observational design precludes causal inference. LASSO regularization helps guard against overfitting, but external validation in independent cohorts is essential before these predictors enter clinical workflows. Still, the integration of intraoperative clot data with routine laboratory values represents a pragmatic, potentially low-cost path toward shrinking the ESUS category — an incremental but genuinely useful advance in precision stroke medicine.