Abstract Number: Esoc2026lt81 Validation of a Trustworthy AI-Based Clinical Decision Support System for Predicting Patient Outcome in Acute Stroke Treatment (Validate)

Abstract Background and aims Artificial intelligence(AI)-based clinical decision support systems(CDSS) can rapidly analyze patient data to support clinical decision-making. Although acute ischemic stroke(AIS) reperfusion treatments benefits are established, using individualized outcome predictions for hyperacute treatment decisions remains challenging. We aimed to test the VALIDATE-CDSS providing real-time, individualized outcome predictions to support treatment selection in AIS. Methods Prospective non-interventional time-motion study of consecutive AIS at three tertiary centers. A shadow researcher collected real-time data via mobile application; clinicians remainded blinded to predictions. For each patient, 3 outcome predictions were locally generated:clinical data alone, +non-contrast CT, and +CT angiography. The system predicts 3-month modified Rankin Scale(mRS) for four treatments: intravenous thrombolysis (IVT), mechanical thrombectomy(EVT), combined IVT+EVT, or no reperfusion(NoREP). Primary outcome: feasibility and usability of VALIDATE-CDSS in real-world hyperacute stroke care. Secondary outcomes: predictive accuracy, treatment concordance, and performance across subgroups. Results 290 AIS patients (53% men, median NIHSS 41-12) were enrolled over 17 months, representing 38% of eligible office-hour strokes. Treatments received: 16.2% IVT, 21.4% EVT, 15.5% both, 46.9% NoREP. All patients had ≥1 prediction; 62.2% had all three. VALIDATE-CDSS achieved excellent System Usability Scale score (82.1/100). However, challenges emerged regarding real-time integration, moderate clinician confidence in predictions, and visualization comprehension. In 28% of cases, neurologists agreed with predictions; only 9% would have changed treatment according to results. Secondary outcomes are being analyzed now. Conclusions AI-based CDSS application in hyperacute stroke is feasible, achieving excellent usability despite challenges in real-time integration and user confidence. Conflict of interest