A multinational collaboration led by Northwestern University and KU Leuven introduces an XGBoost-based clinical decision tool to predict acute kidney injury and survival in neonates treated with therapeutic hypothermia. By integrating gestational age, birth weight, postnatal age, and early serum creatinine trends, the model achieves AUC 0.95 and 75% accuracy on cross-validated multicenter data, enabling timely risk stratification and individualized neonatal management.
Key points
- XGBoost classifier uses gestational age, birth weight, postnatal age, and daily serum creatinine to predict five neonatal outcome classes.
- Trained on 1,149 hypothermia-treated neonates and 801 controls with stratified 10-fold cross-validation and patient-level data splits.
- Achieves mean AUC 0.95 and 75.1% overall accuracy, outperforming existing neonatal AKI biomarkers for early risk stratification.
Why it matters: This high-accuracy AI tool enables clinicians to identify at-risk neonates under therapeutic hypothermia earlier, potentially improving interventions and outcomes.
Q&A
- How does the XGBoost model handle serial creatinine data?
- Why is predicting AKI in cooled neonates challenging?
- What does an AUC of 0.95 signify?
- What is therapeutic hypothermia in neonates?