Systematic review of frailty risk prediction models in stroke patients

JIANG Yanling; LIAO Jianmei; HUANG Niyan

doi:10.3969/j.issn.2096-9058.2025.06.005

Abstract：Objective To systematically evaluate frailty risk prediction models for stroke patients and provide a sci-entific screening tool for healthcare workers.Methods CNKI,Wanfang,VIP,CBM,PubMed,Embase,Web of Sci-ence,and The Cochrane Library databases were systematically searched for studies on frailty risk prediction models for stroke patients,with the search period from database establishment to April 2025.Two researchers independently screened literature and extracted data.The PROBAST tool was used for assessment of bias risk and applicability.RevMan 5.4 software was applied for Meta-analysis of predictors with a frequency>2 times,and MedCalc software was used to eval-uate the performance of the included prediction models.Results A total of 13 studies were included.The AUC of the prediction models ranged from 0.629 to 0.94,and the pooled AUC by MedCalc was 0.857(95%CI:0.796-0.917),in-dicating moderate predictive performance.Meta-analysis results showed that age,depression,falls,dysphagia,diabetes,malnutrition,National Institutes of Health Stroke Scale(NIHSS)score,comorbidity,living alone,activities of daily living(ADL),and physical exercise were significant predictors of frailty in stroke patients(all P<0.05).Conclusions Frail-ty risk prediction models for stroke patients have certain value,but they carry a high risk of bias and their performance needs further improvement.Future studies should adopt machine learning technology and conduct large-sample,multi-cen-ter,prospective research.Healthcare workers should scientifically and reasonably apply the prediction models according to patients'individual differences to improve model accuracy.