Abstract: Objective To systematically evaluate the frailty risk prediction model for maintenance hemodialysis patient.  Methods  PubMed, Embase, CINAHL, Web of Science, Cochrane Library, CNKI, Wanfang, VIP, and SinoMed were systematically searched with a timeframe from establishment of the database to August 29, 2024. Two investigators screened the literature, extracted data, and evaluated bias risk and suitability of the included studies.  Results  Twelve literatures were included, including 16 risk prediction models. Eleven models were validated, and 9 models were calibrated. There were 7～24 candidate variables. Thirteen models reported the area under the curve (AUC) of 0.66～0.998 at the establishment of the models; 4 models reported the AUC of 0.828～0.939 by internal validation, and 2 models reported the AUC of 0.865～0.904 by external validation. There were 4～10 predictors; age, charlson comorbidity index (CCI), gender applicability and depression were the common predictors with the highest frequency. Twelve studies had high risks of bias. Eleven studies showed better applicability to study population, predictors, outcomes and overall subjects.  Conclusions  The frailty risk prediction models for maintenance hemodialysis patients have better predictive performance and clinical value. However, these predictive models have higher bias, limiting their extrapolation. A variety of machine learning algorithms can be used for modeling in the future. We should pay attention to the predictors with higher frequencies in the model and carry out external validation of multiple regions, multiple centers, and large samples for these predictors, to develop predictive models with better prediction performance, more clinical utility, and higher generalized suitability. 

Key words: Maintenance hemodialysis, Frailty, Forecasting model