Identification of key clinical features for pediatric respiratory syncytial virus infection using machine learning

BACKGROUND: Respiratory syncytial virus (RSV) is the leading cause of lower respiratory infections in children. However, current diagnosis currently relies heavily on the clinicians’ subjective judgment, potentially delaying appropriate intervention. Identifying the primary clinical features of RSV in hospitalized pediatric patients can aid triage. Therefore, we analyzed a publicly available dataset comprising 768 pediatric patients, 135 of whom had confirmed RSV infection. METHODS: A binary classification model was developed using the CatBoostClassifier algorithm and evaluated through five-fold cross-validation. Feature importance was assessed using SHapley Additive exPlanations (SHAP) values, and the threshold values for the top features were derived from the dependency plots. RESULTS: The model achieved an average area under the receiver operating characteristic curve (AUC) of 0.770, indicating a moderate predictive performance. The three most important features are weight, respiratory rate, and oxygen saturation (SpO₂). A simplified scoring system based on thresholds for weight (≤ 10 kg), respiratory rate (≥ 50 breaths/min), and SpO₂ (≤ 97%) achieved an AUC of 0.726. CONCLUSIONS: These findings suggest that machine learning can identify clinically meaningful features for RSV prediction in hospitalized children with severe pneumonia. The proposed scoring system may complement conventional, existing scoring approaches. However, prospective validation is required to confirm its generalizability and clinical utility.