Cognitive frailty risk prediction models in patients with chronic kidney disease in China: a systematic review and meta-analysis

Cognitive frailty, defined as the coexistence of physical frailty and cognitive impairment, is common among patients with chronic kidney disease (CKD) and is associated with adverse outcomes such as falls, hospitalization, dementia, and mortality. Because cognitive frailty is potentially reversible, early identification of individuals at high risk is essential. In recent years, several cognitive frailty risk prediction models have been developed for CKD patients, yet their methodological quality, predictive performance, and clinical applicability vary considerably. A systematic evaluation of existing models is therefore needed to summarize their characteristics, assess bias and applicability, and provide evidence-based guidance for future model development and optimization. Databases including CNKI, Wanfang, VIP, CBM, PubMed, Web of Science, Embase, CINAHL and The Cochrane Library were searched up to September 1, 2025. Two researchers independently screened studies and extracted data. The PROBAST tool was used to evaluate risk of bias and applicability, and STATA 14 was applied for meta-analysis of AUC values and predictive factors. Thirteen studies involving 17 prediction models were included. All 13 included studies were conducted in China. Eleven studies performed internal validation and four reported external validation. Overall, the models showed high risk of bias but good applicability. Reported AUC values ranged from 0.791 to 0.990. The pooled AUC was 0.93 (0.90–0.95), indicating strong predictive performance. Significant predictors of cognitive frailty included activities of daily living, depression, social support, education level, age, marital status, nutritional status, hemoglobin, emotional distress, and health empowerment (P < 0.05). Existing risk prediction models for cognitive frailty in patients with chronic kidney disease demonstrate generally good discriminatory ability but remain at an early stage of methodological development. Several aspects require further improvement, including more rigorous predictor selection strategies, standardized data processing procedures, adequate sample sizes, and more robust internal and external validation. In addition, most models are derived from single-center datasets and rely primarily on traditional statistical approaches. Future research should therefore incorporate larger multicenter datasets, strengthen external validation across diverse populations, and explore advanced modeling techniques such as machine learning to improve predictive accuracy, robustness, and generalizability. Notably, all included models were developed in Chinese populations, and their applicability to other ethnic groups and healthcare systems remains uncertain, highlighting the need for validation in more diverse global populations.