Composite small vessel disease scores predict hemorrhagic transformation after thrombectomy: a machine learning study
Thiago Oscar Goulart1, Rui Kleber do Vale Martins-Filho2, Millene Rodrigues Camilo2
1Department of Medicine, Temerty Faculty of Medicine, University of Toronto, Toronto, ON, Canada; Department of Neurology, Sunnybrook Health Sciences Centre, University of Toronto, Toronto, ON, Canada; Department of Epidemiology, Harvard T.H. Chan School of Public Health, Boston, MA, USA; Department of Neuroscience and Behavioral Sciences, Ribeirão Preto Medical School, University of São Paulo, Ribeirão Preto, São Paulo, Brazil.
Background:
Hemorrhagic transformation (HT) is a major complication of acute ischemic stroke (AIS), especially after mechanical thrombectomy (MT) and thrombolysis. This study evaluated whether composite small vessel disease (CSVD) scores-modified Small Vessel Disease (mSVD) and Brain Frailty Score (BFS)-predict HT and symptomatic intracranial hemorrhage (SICH) after MT.
Method:
We conducted a retrospective analysis of a prospectively maintained registry, including AIS patients with anterior circulation large vessel occlusion who underwent MT, with or without Intravenous Thrombolysis (IVT), at a tertiary hospital in a middle-income country. CSVD burden was quantified using mSVD and BFS scores, based on the presence of White Matter Hypodensities (WMH), brain atrophy, and lacunes. Logistic regression and XGBoost assessed predictive performance using AUC and SHapley Additive exPlanation (SHAP) for accuracy and feature importance.
Results:
Among 351 patients (mean age 67.0 ± 7.0 years; 48.3% male; 48.9% received IVT), 18.3% developed parenchymal hematomas (PH-1/PH-2). Inter-rater agreement was robust (mSVD weighted-κ = 0.80; BFS weighted-κ = 0.74). The model with mSVD showed the highest predictive accuracy for HT (AUC = 0.913) and SICH (AUC = 0.862), outperforming BFS (HT:AUC = 0.849; SICH:AUC = 0.839) and individual imaging markers (HT:AUC = 0.864; SICH:AUC = 0.744). IVT was not associated with increased HT risk after adjusting for SVD scores. SHAP analysis identified NIHSS, glycemia, systolic BP, age, and mSVD (mean-SHAP = 0.503) as top predictors; mSVD ranked above BFS (0.369), while individual imaging markers were not among the top features.
Conclusions:
CSVD scores, particularly mSVD score, independently predict the risk of HT after MT, even after accounting for acute clinical variables. These scores may enhance early risk stratification and inform treatment decisions. Validation in multicenter cohorts using advanced imaging is warranted.