Research Article: Risk prediction of pediatric intensive care unit admission in children with respiratory syncytial virus infection using interpretable machine learning

Abstract:
Respiratory syncytial virus (RSV) is a major cause of pediatric acute lower respiratory infection. Early prediction of pediatric intensive care unit (PICU) transfer may support clinical decision-making and resource allocation. We aimed to develop and temporally validate an interpretable machine learning model for predicting PICU admission in hospitalized children with RSV. We included children aged 29 days–18 years with laboratory-confirmed RSV at a single center (January 2023–April 2025). The prediction target was PICU admission within 48 h of hospital admission. Internal validation was performed using random split-sample (80% training / 20% internal test set) and 10-fold cross-validation during hyperparameter tuning. The development cohort (January 2023–March 2025; n = 1,606, 209 PICU admissions) was randomly split into training (80%, n = 1,285, 167 events) and internal test (20%, n = 321, 42 events) sets. A separate temporal external validation cohort (April 2025; n = 94, 12 events) from the same institution was used for temporal external validation. Ten machine learning algorithms were trained using day-one clinical and laboratory variables and evaluated by AUROC, average precision, classification metrics, calibration curves, and decision curve analysis. SHAP was used for interpretation. Among 1,606 children, 209 required PICU admission. Final predictors included dyspnea, serum ferritin, wheezing, immunoglobulin G, interleukin-6, preterm birth, and personal history of wheezing. Random forest performed best among ten algorithms, with AUROC 0.94, average precision 0.87, accuracy 0.95, precision 0.86, recall 0.76, and F1 score 0.81 in the internal test set. In the temporal external validation cohort, the model achieved an AUROC of 0.92, average precision of 0.82, accuracy of 0.95, precision of 0.94, recall of 0.68, and F1 score of 0.79. Calibration plots, decision curve analysis, and SHAP analysis suggested acceptable calibration, potential clinical utility, and interpretability. We developed an interpretable machine learning model to predict PICU transfer in children with RSV. The random forest model performed well, but development estimates may remain optimistic without bootstrap-based optimism correction, and the small temporal external validation cohort limits precision. Further geographical, institutional, and healthcare-system validation is needed before clinical implementation.

Introduction:
Respiratory syncytial virus (RSV) is a major cause of pediatric acute lower respiratory infection. Early prediction of pediatric intensive care unit (PICU) transfer may support clinical decision-making and resource allocation. We aimed to develop and temporally validate an interpretable machine learning model for predicting PICU admission in hospitalized children with RSV.

Read more