Objective To systematically review the construction methods, predictive factors, and performance of postoperative delirium risk prediction models in children, to provide references for the improvement of prediction modelsand clinical practice. Methods Following a scoping review framework, databases including CNKI, Wanfang Data, SinoMed, VIP, PubMed, Embase, Web of Science, and Cochrane Library were searched from the inception to March 2025. Studies reporting predictive models for postoperative delirium in children were included,data were extracted and synthesized. Results Nine studies were included, reported postoperative delirium incidence ranging from 11.10% to 58.68%. Modeling methods comprised multivariate logistic regression analysis and machine learning. Age, surgical type, and pain intensity were the primary predictors of postoperative delirium risk in children. The area under the receiver operating characteristic curve (AUC) for the predictive models developed in the 9 studies ranged from 0.767 to 0.960. Among these, 5 models reported Hosmer-Lemeshow test results, all P>0.05. Conclusion Current research on postoperative delirium risk prediction models in children remains in an exploratory phase, with room for improvement in methodological quality. Future efforts should focus on developing more scientific, precise, and practical prediction models through diverse modeling approaches and rigorous validation to better support clinical decision-making and enhance pediatric nursing care quality.