Abstract
OBJECTIVE: Early differentiation of stroke etiology in acute large vessel occlusion stroke (LVOS) is crucial for optimizing endovascular treatment strategies. This study aimed to develop and validate a prediction model for pre-procedural etiological differentiation based on admission laboratory parameters. METHODS: We conducted a retrospective cohort study at a comprehensive stroke center, enrolling consecutive patients with acute LVOS who underwent endovascular treatment between January 2018 and October 2024. The study cohort (N = 415) was split into training (n = 291) and validation (n = 124) sets using a 7:3 ratio. We applied machine learning techniques-the Boruta algorithm followed by least absolute shrinkage and selection operator regression-for variable selection. The final predictive model was constructed using multivariable logistic regression. Model performance was evaluated through the area under the receiver operating characteristic curve (AUC), calibration plots, and decision curve analysis. We then developed a web-based calculator to facilitate clinical implementation. RESULTS: Of 415 enrolled patients, 199 (48.0%) had cardioembolism (CE). The final model incorporated six independent predictors: age [adjusted odds ratio (aOR) 1.03], male sex (aOR 0.35), white blood cell count (aOR 0.86), platelet-large cell ratio (aOR 1.06), aspartate aminotransferase (aOR 1.02), and non-high-density lipoprotein cholesterol (aOR 0.75). The model demonstrated good discriminatory ability in both the training set (AUC = 0.802) and the validation set (AUC = 0.784). Decision curve analysis demonstrated consistent clinical benefit across threshold probabilities of 20%-75%. CONCLUSION: We developed and internally validated a practical model using routine admission laboratory parameters to differentiate between CE and large artery atherosclerosis in acute LVOS. This readily implementable tool could aid in preoperative decision-making for endovascular intervention.