Abstract 4135852: Safety and Post-operative Complications of Endovascular Versus Surgical Versus Follow-up and Medical Treatment for Patients with Vertebrobasilar Artery Stenosis: Propensity Score Weighting and a Machine Learning Driven Analysis

Background: Vertebrobasilar artery stenosis (VBAS) can cause posterior circulation strokes (PCS). Optimal management is controversial, with options including medical therapy (MT), endovascular stenting (ES), and surgical revascularization (SR). This study compares outcomes of these treatments and evaluates the correlation between clinical features and medical history with 30-day outcome. Methods: Patients with VBAS were identified from the 2017-2018 National Inpatient Sample (NIS). Propensity scores adjusted for baseline differences. Outcomes included mortality, neurological complications (NC), discharge destination (DD), length of stay (LOS), total charges (TC), and procedural complications (PC). Predictive ability of clinical variables was assessed using logistic regression (LR) and machine learning techniques (MLT). Results: Of 1,343 patients, 1,061 (79.0%) received NI, 234 (17.4%) underwent ES, and 24 (1.8%) had SR. Mean age was 69.45 years, with 64.1% male. Demographics: 69.8% White, 14.9% Black, 10.0% Hispanic, and 5.3% other races. Hypertension (HTN, 85.4%) and diabetes (DM, 18.9%) were prevalent. After propensity weighting, ES was associated with higher odds of mortality, surgical/medical complications (SMC), and device/graft complications (DGC) compared to NI. SR showed a non-significant trend toward higher non-home discharges (NHD). ES and SR groups had higher resource utilization with longer LOS and greater TC. Clinical variables alone were weak predictors, with AUC values ranging from 0.454 to 0.71 across different outcomes and models. Conclusion: ES of VBAS was associated with higher mortality and complication rates compared to MT alone, with inconsistent benefits for NC. SR also carried elevated risks without clear advantages over MT. These results support that current clinical independent variables from the NIS are weak predictors. This highlights the limitation of the database in relying solely on clinical and medical history, and suggests that future use of radiological and anatomical features can improve predictions of outcomes and determination of subgroups that can benefit from certain treatment. More studies should be conducted, including post hoc analyses based on radiological and anatomical features, to better inform treatment decisions and determine subgroups that can benefit from intervention or surgery. These findings suggest a need for judicious patient selection and reinforce the role of optimal MT.