Frailty prediction models in older adults with stroke in China: A systematic review and meta-analysis

Frailty is a crucial factor influencing the prognosis of elderly stroke patients. Recent prediction models have focused heavily on predictive performance while often neglecting rigorous methodological quality assessment, thereby limiting their clinical applicability. This systematic review and meta-analysis aims to evaluate the predictive performance and methodological quality of existing frailty prediction models for older stroke patients in China, identify the most frequently included predictors, and provide evidence-based guidance for future research.

A comprehensive literature search was conducted in PubMed, Embase, Web of Science Core Collection, Cochrane Library, EBSCOhost, and several Chinese databases for studies published up to May 2025. Two reviewers independently extracted data using the Checklist for Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modelling Studies and assessed the risk of bias with The Prediction model Risk Of Bias ASsessment Tool. A random-effects meta-analysis was conducted using R software (version 4.5.1) to pool the area under the curve (AUC) values of the prognostic models for frailty.

A total of 12 studies were included in the review, of which 11 were eligible for the meta-analysis. The model AUCs ranged from 0.629 to 0.915. Frequently identified key geriatric predictors included age, neurological function, activities of daily living, physical exercise, and nutritional status. Methodological appraisal indicated a high risk of bias across all models, primarily due to non-prospective designs and flaws in the analysis phase. The pooled AUC from the meta-analysis was 0.82 (95% CI: 0.76 - 0.88, I² = 92.0%, P < 0.0001). Limitations of the included studies were small sample sizes and inappropriate handling of continuous variables.

This systematic review finds that frailty prediction models for older Chinese stroke patients have modest discriminative ability (pooled AUC 0.82). However, all models suffer from methodological flaws, including non-prospective designs, small sample sizes, and a lack of external validation. These limitations undermine their reliability and clinical utility. Rather than endorsing any specific model, this review maps the current evidence and identifies key areas for improvement. Future research should prioritize rigorous development and external validation in prospective studies.