ObjectiveTo systematically evaluate the methodological quality and predictive performance of acute kidney injury (AKI) prediction models following coronary artery bypass grafting (CABG), aiming to identify reliable tools for clinical practice and provide evidence-based guidance for developing higher-quality models in future. MethodsA systematic literature search was conducted across CNKI, Wanfang Data, VIP, SinoMed, PubMed, Web of Science, EMbase, and Cochrane Library databases from inception to October 2025. Two independent reviewers screened studies, extracted data, and performed prediction model risk of bias assessment. Qualitative synthesis was followed by meta-analysis using STATA 15.0 software. ResultsA total of 21 studies involving 55 prediction models were included. The majority of the studies demonstrated good applicability, but exhibited high overall risk of bias. The models showed favorable discriminative ability, with areas under the receiver operating characteristic curves ranging from 0.707 to 0.958 in training cohorts, and a pooled area under the curve of 0.79 [95%CI (0.76, 0.82)]. The area under the receiver operating characteristic curve in the validation set ranged from 0.55 to 0.90, with a pooled area under the curve of 0.80 [95%CI (0.78, 0.81)]. Most models were presented as Nomograms. Common predictors included age, serum creatinine, estimated glomerular filtration rate, hemoglobin, uric acid, cardiopulmonary bypass, and intra-aortic balloon pump. ConclusionCurrent prediction models demonstrate satisfactory discrimination performance but are limited by single-center development, insufficient external validation, and methodological biases. Future multicenter prospective studies should optimize variable processing and model validation strategies to enhance clinical applicability and generalizability of predictive tools.
