Predicting 30-Day Readmission After Stroke: A Systematic Review and Meta-Analysis to Inform Predictor Selection

Background: Thirty-day readmission after stroke remains common, yet contemporary readmission rates, prediction model performance, and predictor domains have not been comprehensively synthesized. Methods: Following PRISMA guidelines, we searched PubMed, Embase, Web of Science, Scopus, and Google Scholar for studies published between 1 January 2021 and 9 October 2025. Readmission proportions and model discrimination, measured by area under the receiver operating characteristic curve (AUC), were pooled using random-effects meta-analysis. Heterogeneity was assessed using I². Predictors were summarized across studies by domain. Results: Twenty studies met inclusion criteria: 15 studies comprising 358,434 patients contributed quantitative data, and 5 were included in qualitative synthesis only. The pooled proportion was 12.9% (95% CI: 10.1-15.8%), with subgroup estimates of 14.2% (95% CI: 11.9-16.6%) for all-cause and 3.6% (95% CI: 0.5-6.7%) for stroke-specific readmissions. Study-level AUCs ranged from 0.59 to 0.88, with a pooled AUC of 0.69 (95% CI: 0.64-0.75), indicating moderate discrimination. Substantial heterogeneity was observed (I² > 98%, p < 0.001), and pooled estimates should be interpreted cautiously. Predictor selection was poorly standardized and largely driven by data availability, with inconsistent inclusion of key clinical and post-discharge domains such as stroke severity, functional status, discharge disposition, post-discharge care, and social determinants of health. Conclusions: Thirty-day readmission after stroke remains common, and currently available models demonstrate modest predictive discrimination; no consistently high-performing, broadly generalizable prediction model has yet emerged. Improving prediction will require broader predictors capturing stroke severity, care transitions, follow-up, and patient context, along with external validation and integration into clinical workflows.