Machine learning-based prediction models for severe

Objectives: This study aimed to systematically evaluate the predictive performance and methodological characteristics of machine learning-based models for predicting progression to severe Methods: A comprehensive literature search was conducted in PubMed, EMBASE, Web of Science, Cochrane Library, CNKI, and Wanfang databases from inception to November 2025 to identify studies developing or validating prediction models for SMPP in children. Data on study characteristics, modeling algorithms, predictors, and performance metrics were extracted. A narrative synthesis was performed to summarize model characteristics, predictors, and modeling approaches, while model discrimination was quantitatively synthesized using pooled area under the receiver operating characteristic curve (AUC). Subgroup analyses were conducted according to modeling algorithms. Methodological quality and risk of bias were assessed using the PROBAST tool. Results: A total of 13 studies were included. The reported prevalence of hypoglycemia ranged from 17 to 33%. The AUC for predictive models ranged from 0.81 to 0.90. Subgroup analyses showed that machine learning-based models such as XGBoost and random forest generally reported higher AUC values compared with other modeling approaches. Commonly reported predictors included age, insulin use, body mass index, HbA1c, creatinine, and history of hypoglycemia. Conclusion: Research on risk prediction models for SMPP in children is still at a developmental stage. Although current models demonstrate high discriminatory performance, methodological limitations and limited clinical translation remain. Future studies should focus on developing robust, interpretable machine learning models and facilitating their integration into pediatric clinical practice. Systematic review registration: https://www.crd.york.ac.uk/PROSPERO/ CRD42020190338.