Abstract 43: Development and Internal Validation of Risk Scores to Predict Survival to Discharge Following Pediatric In-Hospital Cardiac Arrest

Abstract

Introduction: In-hospital cardiac arrest (IHCA) in the pediatric population is associated with poor survival and neurological outcomes. We aimed to develop and internally validate a risk score to predict survival to discharge in pediatric IHCA patients. Methods: For model development and validation, we included pediatric IHCA patients captured in the Get With The Guidelines® registry (provided by the American Heart Association) between 2005 and 2021. We used logistic regression (LR), classification and regression trees (CART), and artificial neural networks (ANN) to develop models using 70% of the data and validate them using the remaining 30% of the data. The discrimination of the models was compared based on the area under the receiver operating characteristic curve (AUC), calibration based on the ratio of observed to expected (O: E), and predictive accuracy based on percent survival in each risk group. Results: We included 6141 pediatric patients (mean age 4.82 years, 41.3% infants, 54.7% male). Most patients survived to hospital discharge (60.9%). We developed separate models for infants vs. older children using LR, CART, and ANN. The least absolute shrinkage and selection operator (LASSO) identified age, illness category, acyanotic cardiac malformation, hepatic insufficiency, hypotension/hypoperfusion, metabolic/electrolyte abnormality, metastatic/hematologic malignancy, renal insufficiency, septicemia, and pediatric cerebral performance score on admission as important predictor variables for analysis. Table 1 shows a comparison of model parameters. Logistic regression models were found to be better at classifying survivors into risk categories with higher specificity and discrimination. Conclusion: Pediatric patients experiencing IHCA can be classified into low, moderate, and high-risk categories using a simple risk score and easily identified pre-arrest variables. These risk scores can support clinical decisions, facilitate research, and help monitor the quality of medical services.