Risk prediction models for frailty in stroke patients: a systematic review and meta-analysis

To systematically review the frailty risk prediction models for stroke patients.

Eight databases including PubMed, Web of Science, the Cochrane Library, Embase, CINAHL, CNKI, Wanfang Database, VIP and SinoMed were systematically searched from the inception of the databases to May 31, 2025.

Studies were screened independently by two researchers with systematic evidence-based training. In this review, We extracted data based on Data Extraction for Systematic Reviews of Prediction Modeling Studies (CHARMS).

Two researchers independently screened the literature and extracted study data. We employed the Prediction Model Risk of Bias Assessment Tool (PROBAST) to assess both the risk of bias and applicability of the included studies.

A total of 19 studies and 24 frailty risk prediction models for stroke patients were included. The area under the Receiver operating Characteristic curve (AUC) of the included models ranged from 0.629 to 0.940, with 21 models having an AUC > 0.7. The combined AUC value of the 10 validation models was 0.87. However, all studies were at high risk of bias according to PROBAST, suggesting this estimate is likely inflated due to methodological weaknesses. The most frequently used predictors in the included models were age, activities of daily living, and NIHSS score.

Current frailty prediction models for stroke patients have methodological weaknesses and high risk of bias. All included studies were from China (2023-2025), severely limiting generalizability to other populations. The pooled AUC (0.87) is likely inflated and should be interpreted cautiously. Rigorous external validation in diverse settings is needed before clinical use. Future studies should expand the sample size, strictly carry out study design and carry out multi-center external validation.

https://www.crd.york.ac.uk/PROSPERO/view/CRD420251075547.