Development and validation of a risk score to predict mortality in the pediatric intensive care unit using routinely collected data in Northwest Ethiopia: a retrospective study

Early prediction of mortality in critically ill patients is essential for informed clinical decision-making, effective risk stratification, and optimal resource allocation. Most existing pediatric-specific mortality prediction models, however, depend on extensive laboratory testing and advanced monitoring and intervention tools, which limit their applicability in resource-limited settings. To develop and internally validate a mortality risk score for pediatric patients admitted to the intensive care unit (ICU). A retrospective cohort study of 565 pediatric patients in northwest Ethiopia was conducted. Data were extracted from medical records using a structured data abstraction tool developed in Kobo Toolbox V2024.2.S. Descriptive statistics, including mean, median, and standard deviation, were summarized and presented in text and tables. The least absolute shrinkage and selection operator (LASSO) was used to select variables for the multivariable logistic regression model. The reduced model’s performance was evaluated using sensitivity, specificity, Brier score, and area under the curve (AUC). Decision curve analysis was conducted to assess clinical utility, and results were presented in a nomogram. The observed mortality rate was 27.3% (95% CI: 23.6%, 30.9%). The final risk score incorporated nine predictors: rural residence, cardiac comorbidity, oxygen saturation (SpO2) at admission, temperature at admission, Glasgow Coma Scale (GCS), urine output, vasopressor use, lowest hemoglobin, and thrombocytopenia. The model demonstrated an area under the receiver operating characteristic curve (AUROC) of 0.86 (95% CI: 0.83, 0.90) and exhibited good calibration. Sensitivity was 85%, specificity was 75%, the negative predictive value was 93.1%, and the positive likelihood ratio was 3.4. Decision curve analysis indicated a greater net benefit compared to the “treat-all” or “treat-none” strategies. A mortality risk score model for the pediatric intensive care unit was developed using readily available clinical profiles. The model demonstrated strong discrimination, calibration, sensitivity, and specificity, supporting its utility for predicting mortality in critically ill children. Following external validation, the model could be implemented in clinical practice and quality improvement programs.