Document resource
Accurate outcome prediction in paediatric solid organ transplantation (SOT) is essential for patient-specific risk stratification. The increasing availability of electronic health records and transplant registries, combined with advances in machine learning (ML), has prompted growing interest in data-driven prognostication. However, evidence supporting ML-based prediction in paediatric SOT remains limited. Therefore, this review aimed to: (1) identify and evaluate ML models developed for predicting outcomes in paediatric kidney, lung and heart recipients, and (2) to assess model quality, risk of bias, and applicability using the updated PROBAST-AI criteria.A systematic search was conducted in February 2025. Studies were eligible if they applied ML to predict post-transplant outcomes in SOT recipients. Studies using ML for phenotypic clustering or diagnostic purposes were excluded. Data extraction included both model details and clinical context. Model quality, risk of bias, and applicability were assessed using PROBAST+AI, which comprises 34 signalling questions across participants, data sources, predictors, outcomes, and analysis.Of 453 records screened, 12 met the inclusion criteria. Sample sizes ranged from 68 to 8,349 patients, with larger datasets sourced from national transplant registries. Frequently used algorithms included random forest, gradient boosting, support vector machines, and neural networks. Random forest outperformed other models in six of the seven studies where it was evaluated. PROBAST-AI found high or unclear risk of bias and quality concerns in most studies, mainly in the analysis domain. Key issues included small datasets, univariate feature selection and inconsistent handling of missing data. Only two studies shared code publicly and no study conducted external validation.This is the first systematic review to evaluate ML-based outcome prediction in paediatric SOT. While these studies support proof-of-concept for ML in paediatric prognostication, future efforts must prioritise externally validated models co-developed by clinicians and data scientists to support safe and effective integration into clinical practice.Acknowledgements for Funding or Support CB is funded by GOSH Children’s Charity Award. This work is supported by the NIHR GOSH BRC. The views expressed are those of the author(s) and not necessarily those of the NHS, the NIHR or the Department of Health.