Interpretable machine learning–based prediction of 90-day mortality after endovascular treatment in acute large vessel occlusion stroke

Vollständiger Abstract

Worum geht es in dieser Arbeit?

Background Despite high recanalization rates after endovascular treatment (EVT) for acute ischemic stroke due to large vessel occlusion (AIS-LVO), 15%−20% of patients still die within 90 days. Existing prognostic models predominantly target functional disability rather than mortality, and rarely incorporate calibration, clinical utility, or interpretability. We aimed to develop and internally validate interpretable machine learning (ML) models for dynamic post-EVT in-hospital prediction of 90-day all-cause mortality in AIS-LVO patients with successful recanalization. Methods We retrospectively enrolled 449 AIS-LVO patients with successful EVT (modified Thrombolysis in Cerebral Infarction grade 2b−3) at a nationally certified Advanced Stroke Center in Guangxi, southwestern China, between January 2022 and May 2025. Candidate predictors included baseline characteristics, admission laboratory indicators, procedural variables, and post-procedural complications available during hospitalization. Patients were randomly split 8:2 into training ( n = 359) and independent test ( n = 90) sets, stratified by 90-day mortality. Fifteen ML algorithms were optimized via Bayesian hyperparameter search with 5-fold cross-validation, and the top five underwent bootstrap optimism correction. Performance was evaluated across discrimination, calibration, clinical utility (decision curve analysis), and reclassification (NRI/IDI). SHapley Additive exPlanations (SHAP) and permutation importance were applied for interpretability. Results The 90-day mortality rate was 19.6% (88/449). Random Forest achieved favorable overall test-set performance (AUC = 0.806; Brier score = 0.128; Hosmer–Lemeshow p = 0.315), with positive net benefit across threshold probabilities of 0.10–0.60. SHAP analysis identified stroke-associated pneumonia, D-dimer, age, acute renal insufficiency, fasting blood glucose, NIHSS score, brain herniation, and symptomatic intracranial hemorrhage as top contributors, concordant with permutation importance. Compared with LR-LASSO, Random Forest showed comparable discrimination without significant reclassification improvement (NRI = −0.472; IDI = −0.067). Conclusions A Random Forest-based ML model with an online calculator and SHAP-based interpretability supports dynamic in-hospital post-EVT mortality risk stratification, with complication predictors incorporated only after they clinically occur. Prospective external validation in geographically diverse multicenter cohorts is warranted.