Document resource
Background Predicting chronic kidney disease (CKD) progression in paediatric patients remains a major challenge, limiting clinicians’ ability to prognosticate and prepare the child and family for kidney replacement therapy (KRT). While several predictive models have been developed for adults with CKD, comparable tools for children remain scarce. Using machine learning on a large UK tertiary paediatric cohort, we developed a model predicting kidney failure, defined as an estimated glomerular filtration rate (eGFR) below 15 ml/min per 1.73 m 2 or initiation of KRT, before transition to adult care.Methods We used electronic health records from patients coded with ‘ICD N18’ at Great Ormond Street Hospital from 01/01/2001 to 31/01/2025. 34 clinically relevant features for CKD progression were selected, and imputation was undertaken where clinically warranted. Supervised classification models, including Random Forest and eXtreme Gradient Boosting (XGBoost), were trained on data from early CKD stages (eGFR ≥ 45).Results 691 children (median age 8 [IQR 11] years, 41% girls) were followed for a median of 9 [IQR 11] years. The XGBoost model displayed the best performance with a F1-score of 87% and a receiver operating characteristic area under curve (ROC AUC) score of 87% compared to other models trained (shown in table 1). SHAP analysis highlighted age, sodium and systolic blood pressure as the key influential features. The model demonstrated strongest performance when eGFR was ≥ 90 (F1-score 74%) and in younger children between 0-4 (F1-score 94%).Discussion/Conclusion Our study shows that leveraging electronic health records data enables strong machine-learning prediction of kidney failure in children with CKD. The model requires validation against an independent CKD cohort and acceptance by clinicians, before integration into clinical practice.Acknowledgements for Funding or Support This work has been made possible through the ongoing work of the Digital Research Environment team at GOSH DRIVE to provide an on-demand, fully configurable data provision service. Many thanks to our Clinical leads. This activity is part of a collaborative working agreement between Great Ormond Street Hospital NHS Foundation Trust and Roche Products Ltd. M-GB-00025078 | November 2025.Abstract 122 Table 1 Model Accuracy Precision Recall F1 ROC AUC Decision Tree Classifier 0.84 0.84 0.84 0.84 0.78 Random Forest Classifier 0.86 0.86 0.86 0.86 0.87 XGBoost 0.87 0.88 0.87 0.87 0.87