Related Experiment Video
Updated: Jul 22, 2026
06:01
A Thrombotic Stroke Model Based On Transient Cerebral Hypoxia-ischemia
Published on: August 18, 2015
15.4K
Risk prediction models for discharge disposition in patients with stroke: a systematic review and meta-analysis
Chaoran Xu1,2, Lijun Xiang1, Yansi Luo1
1Department of Neurology, Nanfang Hospital, Southern Medical University, Guangzhou, China.
Frontiers in Neurology
|October 23, 2025
Summary
This study evaluated multivariate prediction models for stroke patient discharge needs. While models show good predictive performance (AUC 0.80), they have a high risk of bias and limited clinical application.
Area of Science:
- Medical Informatics
- Health Services Research
- Stroke Medicine
Background:
- Multivariate prediction models are crucial for estimating the need for higher care levels in discharged stroke patients.
- Systematic evaluation and meta-analysis are essential to determine the performance of these predictive models.
Purpose of the Study:
- To systematically evaluate and perform a meta-analysis of multivariate prediction models for estimating the risk of discharged stroke patients needing a higher level of care.
- To assess the performance and identify limitations of existing prediction models in stroke patient discharge disposition.
Main Methods:
- A comprehensive search of multiple databases (CNKI, Wanfang, VIP, SinoMed, PubMed, Web of Science, CINAHL, Embase) was conducted up to September 30, 2024.
- Data extraction and risk of bias assessment using the Prediction Model Risk of Bias Assessment Tool (PROBAST) were performed independently by multiple reviewers.
- Statistical analyses, including meta-analysis, were conducted using Stata 17.0.
Main Results:
- Out of 4,059 retrieved studies, 14 studies comprising 22 models were included. The incidence of non-home discharge varied widely (15-84.9%).
- Common predictors included age, National Institutes of Health Stroke Scale (NIHSS) score, and Functional Independence Measure (FIM) scores. Reported AUCs ranged from 0.75 to 0.95.
- A meta-analysis of five validation models yielded an AUC of 0.80 (95% CI: 0.75-0.86), despite a high overall risk of bias in the included studies due to data sources and reporting.
Conclusions:
- Risk prediction models for stroke patient discharge are in early development, characterized by high bias and low clinical utility.
- Despite limitations, these models demonstrate good predictive performance.
- Future research should prioritize developing interpretable, high-performance machine learning models with enhanced external validation and a focus on clinical application.
