Machine Learning Models for Predicting ICU Admission and Mortality in Pediatric Severe Clinical Pneumonia: A Cohort Study

Abstract Purpose Pneumonia remains a leading cause of morbidity and mortality among children worldwide, especially in low- and middle-income countries. This study explores the application of machine learning (ML) models to predict ICU admission and mortality in pediatric patients with severe pneumonia. Methods This study utilized a dataset, comprising 801 patients with acute respiratory presentation at the Children's Hospital of Rabat, Morocco. Patients meeting WHO criteria for severe pneumonia were included, excluding those who left against medical advice, resulting in 699 patients. Four ML algorithms were employed. Feature selection was performed to identify the top five features influencing ICU admission and mortality. Results The support vector model had a positive likelihood ratio (PLR) of 9.333 (95% CI: 5.181–16.810) and a negative likelihood ratio (NLR) of 0.309 (95% CI: 0.120–0.794), the logistic regression model showed a PLR of 9.625 (95% CI: 5.590–16.574) and an NLR of 0.233 (95% CI: 0.086–0.637), the Gaussian naive Bayes model had a PLR of 7.000 (95% CI: 3.886–12.608) and an NLR of 0.393 (95% CI: 0.194–0.795), and the XGBoost model demonstrated a PLR of 32.84 (95% CI: 8.333–129.47) and an NLR of 0.508 (95% CI: 0.250–1.030) for predicting ICU admission. For predicting mortality, the support vector model showed a lower accuracy. Cyanosis was the most important factor for both ICU admission and mortality. Conclusion ML models demonstrate high accuracy in predicting ICU admission but lower reliability for mortality prediction.