Risk prediction of pediatric intensive care unit admission in children with respiratory syncytial virus infection using interpretable machine learning - Report - MDSpire
Advertisement
Risk prediction of pediatric intensive care unit admission in children with respiratory syncytial virus infection using interpretable machine learning
Clinical Report: Predicting Pediatric Intensive Care Unit Admissions for RSV
Overview
This study developed and validated an interpretable machine learning model to predict PICU admissions in children with RSV. The model demonstrated high accuracy based on initial clinical and laboratory data.
Background
Respiratory syncytial virus (RSV) is a leading cause of pediatric acute lower respiratory infections, with significant hospitalization rates. Early identification of children at risk for PICU admission is critical for effective resource management, especially during peak RSV seasons when healthcare resources are strained. Current predictive models are insufficient for guiding clinical decisions in real-time settings.
Data Highlights
Metric
Internal Test Set
External Validation Cohort
AUROC
0.94
0.92
Average Precision
0.87
0.82
Accuracy
0.95
0.95
Precision
0.86
0.94
Recall
0.76
0.68
F1 Score
0.81
0.79
Key Findings
The model identified key predictors for PICU admission including dyspnea, serum ferritin, and interleukin-6.
Random forest algorithm outperformed other machine learning models with an AUROC of 0.94 in the internal test set.
In external validation, the model maintained high accuracy with an AUROC of 0.92.
SHAP analysis provided interpretability, enhancing understanding of the model's predictions.
Further validation across different healthcare settings is necessary.
Clinical Implications
Understanding the predictors can help prioritize resources and interventions during peak RSV seasons.
Conclusion
The study presents a machine learning approach for predicting PICU admissions in RSV-infected children, highlighting the need for further validation.
