AUTHOR=Zhang Jiali , Fu Yijie , Liu Yan , Liu TianHeng , Deng Yue , Dai LiFei , Zhu Tianmin , Li Hui 

TITLE=Risk prediction models for short-term mortality in ICU stroke patients: a systematic review and meta-analysis

JOURNAL=Frontiers in Neurology

VOLUME=Volume 16 - 2025

YEAR=2025

URL=https://www.frontiersin.org/journals/neurology/articles/10.3389/fneur.2025.1623645

DOI=10.3389/fneur.2025.1623645

ISSN=1664-2295

ABSTRACT=ObjectivesThis study aims to systematically review and evaluate risk prediction models for short-term mortality in ICU stroke patients, thereby providing scientific evidence to inform future model development and clinical application.MethodsWe searched the Cochrane Library, EMBASE, PubMed, and Web of Science for studies on prediction models for short-term mortality in ICU stroke patients, covering the period from January 2005 to January 2025. Data extracted included study characteristics and detailed information on the prediction models. The Risk of Bias and applicability of the models were evaluated using the Prediction Model Risk of Bias Assessment Tool (PROBAST). A meta-analysis was performed using a random-effects model in Stata 18.0, and heterogeneity across studies was assessed using the I2 statistic. Subgroup analyses were conducted based on stroke type, geographic region, and modeling approach. a sensitivity analysis performed to evaluate the robustness of the findings.ResultsA total of 6,874 studies were retrieved, and 12 studies met the inclusion criteria, yielding 14 prediction models, as two studies included two models each that were extracted separately. Four models were externally validated. The reported area under the curve (AUC) values ranged from 0.761 to 0.977. Meta-analysis yielded a pooled AUC of 0.82 (95% CI: 0.80–0.85), indicating good discriminative ability of the models in predicting short-term mortality in ICU stroke patients. However, heterogeneity was high (I2 = 80.1%, p = 0.000). Subgroup analyses by stroke type, modeling approach, and geographical region revealed no statistically significant sources of heterogeneity. The PROBAST assessment shows that all models exhibit high risk of bias and low applicability. The most frequently reported predictors were Glasgow Coma Scale (GCS), white blood cell count (WBC), age, and blood glucose levels.ConclusionThis study shows that prediction models for short-term mortality in ICU stroke patients have good discriminatory performance. However, due to high bias risk and low applicability, their overall quality remains suboptimal. Important predictors such as GCS, WBC, age, and blood glucose levels should be included in future models. Future research should focus on prospective, multicenter, and externally validated studies guided by the PROBAST tool to improve clinical applicability and reliability.Systematic review registrationhttps://www.crd.york.ac.uk/PROSPERO/recorddashboard.