Abstract Number: Esoc2026a454 Predictive Value of Machine Learning for Post-Stroke Mortality Risk: A Systematic Review and Meta-Analysis

Abstract Background and aims With growing attention paid to machine learning (ML) in stroke care, some researchers have investigated the effectiveness of ML in predicting the mortality risk in stroke. However, systematic evidence is still lacking for its effectiveness. Therefore, we conducted this meta-analysis of the value of ML in predicting the stroke mortality risk. Methods A search was made in Cochrane Library, PubMed, Embase, and Web of Science up to June 23, 2025. Pooled risk ratios with 95% confidence intervals (CIs) were derived using the Hartung-Knapp-Sidik-Jonkman method under a random-effects model. Results Sixty-eight studies were included For predicting in-hospital mortality, the pooled C-index was 0.788 (95%CI: 0.766-0.810), with sensitivity and specificity of 0.69 (95%CI: 0.65-0.73) and 0.79 (95%CI: 0.76-0.82), in the validation set. For predicting out-of-hospital mortality, the validation set had a pooled C-index of 0.812 (95%CI: 0.798-0.826), with sensitivity and specificity of 0.68 (0.63-0.72) and 0.82 (0.78-0.85). The logistic regression (LR) model had a C-index of 0.802 (95%CI: 0.762-0.844) and 0.817 (95%CI: 0.793-0.841), respectively, for predicting in-hospital and out-of-hospital mortality. The meta-regression revealed a gradual decline in the predictive performance of the overall model and LR model, and a robust performance of the random forest model. Age, NIHSS score, and stroke-related complications were the most frequently used variables for modeling. Conclusions ML-based prediction for the stroke mortality risk is feasible, and clinical scoring tools can be developed to offer an evidence-based basis for specific prediction protocols for high-risk populations. The findings can contribute to clinical decision-making and resource allocation. Conflict of interest Yujie Chen & Muke Zhou: nothing to disclose