A51 Machine learning-based prediction of stroke outcome: value of immediate vs. 24-hour clinical parameters – insights from large-scale multicenter registry data

<h3>Introduction</h3> Ischemic stroke remains a leading cause of morbidity and mortality, despite substantial progress in acute interventions. Understanding prognostic factors is essential for individualized treatment strategies and optimizing clinical decision-making. <h3>Aim of Study</h3> This study aims to identify and compare clinical parameters available immediately after endovascular treatment and at 24 hours that predict good functional outcome, defined as a modified-Rankin-Scale (mRS) score of 0–2 at 90 days. Analyses are based on a large, multicenter stroke registry and employ machine learning-backed model testing and validation to ensure robustness and generalizability of results. <h3>Method</h3> All patients from the German Stroke Registry were screened (n=18,069; 2015–2023). Patients with anterior circulation stroke and documented 90-day mRS scores were included (n=9,449). Missing data were addressed using Multiple Imputation by Chained Equations (five imputations, 20 iterations). Predictor selection was performed using logistic regression with nested cross-validation (outer: 5-fold, 3 repeats; inner: 5-fold). <h3>Results</h3> Key predictors of good outcome immediately post-procedure included pre-stroke mRS (normalized adjusted odds ratio [naOR] 0.42; 95%CI:0.38–0.46), age (naOR 0.48; 95%CI:0.44–0.52), and successful recanalization (mTICI≥2b; naOR 1.78; 95%CI:1.57–1.83), with AUC of 0.83 (95%CI:0.82–0.85). At 24 hours, NIHSS was the strongest predictor (naOR 0.18; 95%CI:0.16–0.20), improving the AUC to 0.90 (95%CI:0.89–0.91). <h3>Conclusion</h3> Clinical variables available post-procedure, particularly age, pre-stroke disability, and recanalization success, provide strong early prognostic value. Incorporating 24-hour NIHSS, which reflects both immediate treatment effect and early neurological evolution, significantly enhances prediction of functional outcomes. <h3>Conflict of Interest</h3> Yes Helge Kniep and Fabian Flottmann are consultants for Eppdata GmbH. Helge Kniep is shareholder of Eppdata GmbH. Götz Thomalla received fees as consultant from Acandis, Boehringer Ingelheim, Bayer, and Portola, and fees as lecturer from Acandis, Alexion, Amarin, Bayer, Boehringer-Ingelheim, BMS/Pfizer, Daiichii Sankyo and Portola. He serves in the board of the TEA Stroke Study and of ESO. Jens Fiehler is consultant for Cerenovus, Medtronic, Microvention, Penumbra, Phenox, Roche and Tonbridge. He serves in the advisory board of Stryker and Phenox. He is stock holder of Tegus Medical, Eppdata and Vastrax.