Machine learning models show high diagnostic accuracy for predicting pediatric cardiac surgery associated acute kidney injuryMachine Learning Models Show Promise Predicting Pediatric Kidney Injury

BackgroundAcute kidney injury (AKI) occurs in up to 42% of pediatric cardiac surgeries and is associated with prolonged intensive care, increased morbidity and in-hospital mortality. Machine learning (ML) has emerged as a promising approach for early AKI risk stratification by modeling complex, high-dimensional clinical data.ObjectivesTo systematically review and meta-analyze the diagnostic accuracy of ML models for predicting pediatric cardiac surgery associated AKI (CSA-AKI).MethodsWe performed a systematic search of PubMed, ScienceDirect, Springer, and DOAJ. The review protocol was prospectively registered in PROSPERO (CRD420251145645). Study quality was assessed using QUADAS-2 tool and PROBAST + AI. A bivariate random-effects diagnostic meta-analysis was performed using Stata 17.0 to estimate the pooled area under the summary receiver operating characteristic curve (SROC AUC), sensitivity, specificity, likelihood ratios, and diagnostic odds ratio (DOR).ResultsA meta-analysis of seven studies yielded a pooled SROC AUC of 0.91 (95% CI 0.88–0.93), driven predominantly by internally validated models (AUC 0.93, Sensitivity of 0.84, Specificity of 0.95). Externally validated models showed substantially lower performance (Sensitivity 0.70, Specificity 0.80), representing the more clinically relevant benchmark. A sensitivity analysis using median-performing models confirmed directional consistency (AUC 0.85, Sensitivity 0.75, Specificity 0.91). Substantial heterogeneity was observed (I2 = 81.48%).ConclusionML models show promising accuracy for predicting pediatric CSA-AKI. Substantial heterogeneity and limited external validation warrant cautious interpretation and further multicenter validation before clinical use.Systematic Review Registrationhttps://www.crd.york.ac.uk/PROSPERO/view/CRD420251145645.