ABSTRACT NUMBER: ESOC2026A2453 VERITAS NEXUS: PERSONALIZING SELECTION FOR ENDOVASCULAR THERAPY IN VERTEBROBASILAR OCCLUSION VIA MACHINE LEARNING-DERIVED INDIVIDUAL TREATMENT EFFECTS

Abstract Background and aims While the efficacy of endovascular therapy (EVT) for vertebrobasilar artery occlusion (VBAO) is established, the magnitude of individual benefit remains highly variable. Average treatment effects often mask patients who may derive exceptional benefit or those for whom intervention may be futile. We developed VERITAS Nexus, a causal ML framework designed to shift the paradigm from population-level averages to Individualized Conditional Average Treatment Effects (CATE). Methods We performed a pooled analysis of individual patient data from four landmark RCTs (BASICS, BEST, ATTENTION, BAOCHE). Utilizing an ensemble of causal ML architectures, we conducted an as-treated analysis to model counterfactual outcomes. The primary endpoint was the individual probability of a favorable outcome (mRS 0-3 at 90 days) under both EVT and medical management (MM). Results Among the 988 patients analyzed (EVT: 574; MM: 414), the median age was 67 years (IQR 58–74) and median NIHSS was 22 (IQR 13–35). VERITAS Nexus revealed a spectrum of individualized responses, with CATE ranging from negligible to highly significant (mean 13.0%). Individualized benefit profiling showed that 62.6% derived moderate benefit (>5%) and 44.9% derived strong benefit (>15%). The model-driven NNT was 3.1, while the outcome models achieved an AUC of 0.75(95% CI0.694–0.801)for MM and 0.68(95% CI 0.635–0.725)for EVT. Conclusions VERITAS Nexus demonstrates that BAO treatment response is individualistic. By quantifying the specific "benefit margin" for each patient, this tool moves beyond binary "treat vs. no-treat" logic. It provides a robust clinical decision support framework to facilitate tailored intervention and high-precision shared decision-making. Conflict of interest Nothing to disclose