P2.070: Evaluating the Use of Risk Assessment Tools to Predict Hospital-to-Hospital Transfer Patients of Potential Escalation of Care to the Pediatric Intensive Care Unit

Background: Children transported to a pediatric hospital who experience clinical deterioration events that require escalation of care to the intensive care unit (ICU) are at risk for significant morbidity and mortality.1-8 Accurate disposition of hospital-to-hospital transfer patients is important for improving morbidity and mortality outcomes and for effective resource utilization.9 Transport-specific pediatric risk assessment scores that aid hospital-to-hospital patient disposition include subjective variables, resulting in inter-user variability and suboptimal performance.10-15 Objectively derived machine learning models have demonstrated strong predictive capabilities, outperforming traditional severity indices in accurate patient triage and clinical outcomes.16-19 We aimed to test the accuracy of a modified version of pCART, a validated electronic healthcare record (EHR)-based in-hospital risk stratification model, as a disposition tool for hospital-to-hospital pediatric transfer patients9-10. Secondarily we tested modified pCART against previously published transport-risk assessment scores: Transport Risk Assessment in Pediatrics (TRAP) and Transport Pediatric Early Warning Scores (T-PEWS).

Methods: This is a retrospective study utilizing hospital-to-hospital pediatric transfer patients (age < 18 years) admitted to the general pediatric ward at The University of Chicago Comer Children’s Hospital from 2009 to 2023. Using Area Under the receiver operating characteristic Curve (AUC) as our primary metric, we evaluated the performance of a modified pCART, TRAP, and T-PEWS in predicting the likelihood that hospital-to-hospital transfer patients would, within 12 hours of admission, be transferred from the general pediatric ward to the pediatric ICU. All models were modified to include only objective variables: vital signs, Glasgow coma scale, and fractional inspired oxygen.

Results: A total of 7,621 patients were identified as hospital-to-hospital pediatric transfers admitted to the general pediatric floor, of which 220 patients (2.89%) required transfer to the ICU within 12 hours. This modified pCART was able to predict in-hospital transfers from the general pediatric floor to the ICU with modest performance (AUC 0.7048, 95% CI 0.6653-0.7442). Modified pCART was comparable to both a modified TRAP (AUC 0.7145, 95% CI 0.6775-0.7514) and a modified T-PEWS (AUC 0.7083, 95% CI 0.6702-0.7464) in predictive capability.

Conclusion: Prediction models with objective data, including a modified pCART, demonstrated ability to identify hospital-to-hospital transfer patients requiring escalation of care to the ICU. However, improved accuracy is warranted to optimize clinical application of these transport-risk assessment scores. The development of a model that uses solely objective pre-hospital and admission-level data to correctly triage a transported patient has the potential to improve patient outcomes and hospital resource utilization.