Clinical Prediction Models for Acute Kidney Injury in Neonatology-A Systematic Review and Modelling Analysis

Acute kidney injury (AKI) affects up to one-third of neonates admitted to neonatal intensive care units (NICUs). NICUs are highly data-rich environments, offering opportunities to develop clinical prediction rules (CPRs) that use clinical and demographic data to estimate AKI risk. We sought to identify and critically appraise existing CPRs for predicting neonatal AKI.

A systematic search of PubMed and OVID-MEDLINE was conducted for literature published from 1946 to Q2-2025. Eligible studies reported the development or validation of CPRs for neonatal AKI in NICU populations. The review followed PRISMA guidelines and was prospectively registered with PROSPERO (CRD420250653606). Data extraction followed the CHARMS checklist, and risk of bias was assessed using the PROBAST tool.

From 685 identified articles, 11 underwent full-text review and three met inclusion criteria. Sample sizes ranged from 276 to 706 participants. One CPR was derived from multi-centre data; two used prospectively collected datasets. The number of predictor variables ranged between 4 and 10. Internal validation methods varied, with those who reported commenting on apparent performance only and none performing external validation. Hosmer-Lemeshow testing and calibration plots were the most commonly used forms of analysis.

Current CPRs for neonatal AKI show potential but remain underdeveloped and poorly validated. Rigorous research, including external validation prior to integration into clinical workflows, is essential to enable effective early detection of neonatal AKI.