Stroke-associated sarcopenia prediction models show AUC values between 0.681 and 0.970 across various biomarkersNew models help predict muscle loss after a stroke

ObjectiveTo systematically map the study designs, model development methods, predictive performance, and predictors of risk prediction models for sarcopenia in stroke patients, and to identify current research status, gaps, and future directions in this field.MethodsThis review followed the Arksey and O’Malley five-stage framework. Nine databases (PubMed, Web of Science, Cochrane Library, Embase, CINAHL, CNKI, Wanfang, VIP and SinoMed) were searched from inception to October 2025. Two reviewers independently screened and extracted data using a form based on the JBI Manual, with methodological classification guided by the Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD) statement. A narrative synthesis approach was adopted.ResultsA total of 12 studies (2022–2025) comprising 15 prediction models were included, with sample sizes ranging from 80 to 1,226 and stroke-associated sarcopenia (SAS) prevalence rates of 15.8–65.8%. Nine studies were conducted in China, and the most common study designs were case–control and cross-sectional studies. Most studies employed parametric methods, predominantly logistic regression, whereas two studies also employed non-parametric machine learning methods, including decision trees, random forest, and XGBoost. And only one study performed true external validation. Seven studies of multivariable diagnostic prediction models reported AUC values ranging from 0.743 to 0.970, and four studies evaluating the diagnostic accuracy of single or few biomarkers reported AUC values ranging from 0.681 to 0.830. Calibration metrics were inconsistently reported across studies. Predictors spanned six dimensions: demographic and lifestyle factors, muscle morphology and body composition, medical history and comorbidities, neurological and cognitive-psychological factors, nutritional and feeding-related factors, and laboratory indicators.ConclusionCurrent SAS risk prediction models are predominantly single-center, small-sample studies that generally lack external validation and calibration reporting, limiting their clinical translational value. Although previous systematic reviews have addressed SAS prediction models, this review also summarized exploratory single-biomarker studies, including those evaluating inflammatory-immune markers such as CAR, NLR, and HMGB1. These markers may provide new directions for SAS risk prediction; however, each is supported by only one small study and should be considered promising predictors rather than validated predictors. Future research should conduct multicenter, prospective studies using standardized sarcopenia diagnostic criteria, adhere to the TRIPOD statement to construct multidimensional prediction models with external validation, and facilitate the translation of risk screening into clinical practice.Systematic review registrationhttps://doi.org/10.17605/OSF.IO/BWKMR.