RAMA: implementing Machine Learning to develop mortality risk prediction models for NICU patients with Acute Kidney...

Area of Science:

• Neonatal medicine
• Machine learning in healthcare
• Pediatric nephrology

Background:

• Acute Kidney Injury (AKI) is prevalent in Neonatal Intensive Care Units (NICUs).
• Neonatal AKI is linked to severe comorbidities and increased mortality risk.
• Predicting mortality risk alongside AKI severity can inform treatment strategies.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting mortality risk in neonates with AKI.
• To identify the most accurate and interpretable model for this prediction task.
• To create a tool that assists clinicians in managing neonatal AKI cases.

Main Methods:

• Two independent neonatal patient datasets were combined and preprocessed.
• Seven machine learning algorithms were trained and tested, with feature selection applied.
• Models were fine-tuned using Bayesian search to optimize performance and interpretability.

Main Results:

• Tree-based models, including Random Forest, XGBoost, and LightGBM, demonstrated high accuracy and interpretability.
• The RAMA model, based on XGBoost, achieved an AUC-ROC of 0.882 ± 0.132.
• RAMA showed strong predictive performance with an Accuracy of 0.878 ± 0.052 and F1 Score of 0.923 ± 0.029.

Conclusions:

• The RAMA model, utilizing a tree-based algorithm, effectively predicts mortality risk in neonates with AKI.
• This tool can aid clinicians in identifying high-risk neonates for targeted interventions.
• Accurate mortality risk assessment is crucial for optimizing care in neonatal AKI patients.