ABSTRACT NUMBER: ESOC2026A2230 TRIAL-TIME PRECISION: REAL-TIME ADAPTIVE ENRICHMENT AS THE NEXT STANDARD FOR ACUTE STROKE TRIALS

Abstract Background and aims Historically, acute stroke trials often fail because conventional trial designs struggle to manage patient heterogeneity and the extreme urgency of intervention. While traditional designs apply fixed eligibility that can dilute treatment effects, current adaptive designs typically implement "batch" updates at the population level, which fails to optimize care for the individual patient at the time of screening. Methods We propose a new paradigm; real-time adaptive enrichment in which a validated counterfactual model acts as a patient-level "gate" during screening. The system calculates the conditional average treatment effect (CATE) for the specific patient; randomization only proceeds if the predicted absolute benefit exceeds a pre-specified threshold. This threshold is finalized through simulations to balance statistical power, type I error, and fairness. Results The DUSK tool for distal occlusions serves as a proof of concept. This interaction model estimates patient-specific probabilities under endovascular therapy versus medical management, incorporating potential modifiers such as baseline severity. The architecture is designed to integrate into existing platforms which already support complex adaptive randomization. Adoption follows a four-stage roadmap; locking the model for regulatory audit, a passive "shadow mode" validation, a prospective pilot within established platform trials, and eventual scaling to standalone studies. This approach remains fully compatible with reporting standards. Furthermore, it aligns with ethical equipoise by reducing the likelihood of futile patients' allocating. Conclusions Real-time adaptive enrichment transforms trial-time into a precision instrument. By concentrating randomization on predicted responders, this framework accelerates evidence generation for neuroprotection, rescue stenting, and other acute neurological emergencies where conventional designs often fail. Conflict of interest Mohamed F Doheim: Nothing to disclose; Raul Nogueira is the PI of DUSK trial; IIT supported by Stryker.