P1.048: Comparing the Predictive Ability of Neighborhood Disparity Metrics for Pediatric Trauma Mechanisms and Outcomes

Background: Trauma is the leading cause of childhood mortality. We previously demonstrated neighborhood disparity impacts the mechanisms and outcomes of pediatric trauma. Multiple composite neighborhood disparity metrics (NDM) have been created, including Area Deprivation Index (ADI), Social Vulnerability Index (SVI), and Childhood Opportunity Index (COI). Little research has compared these metrics in pediatric trauma. We hypothesized one metric would be a better predictor of pediatric trauma mechanisms and outcomes.

Methods: We compared the NDMs’ predictive ability for pediatric trauma mechanisms and outcomes utilizing data from injured children (≤ 18 years) at a single Level 1 Pediatric Trauma Center from 2016-2021. Outcomes included traumatic brain injury (TBI), Intensive Care Unit (ICU) length of stay (LOS) ≥ 3 days, intubation, ICU admission, blood transfusion within 4 hours of arrival, and Need for Trauma Intervention (NFTI) criteria. Mechanisms included motor vehicle crashes (MVC), non-accidental trauma (NAT) and sports. Patients were excluded for missing home address and inability to calculate NDM. Unadjusted logistic regression models were fit to examine the association between the NDMs and trauma mechanisms and outcomes. These models were used to compare diagnostic accuracy of NDMs with significant associations for ≥ 2 NDMs (p < 0.05) with variables of interest. Empirical receiver-operating characteristic (ROC) curves for each NDM were produced and area under the curve (AUC) and 95% confidence limits were estimated and compared using a contrast test. For models with a significant difference in unadjusted AUCs (p < 0.05), the same method was employed to compare the AUCs of models adjusting for race (white vs. other), ethnicity, insurance status (private vs. other) and age.

Results: Unadjusted models revealed a difference in AUC for the following outcomes: TBI, ICU stay ≥ 3 days, intubation, ICU admission, blood transfusion within 4 hours of admission, and presence of NFTI criteria and for MVC, NAT, and sports. The AUC was highest for models using ADI for TBI, intubation, ICU admission, and NAT. The AUC was highest for the models using COI for ICU LOS ≥ 3 days, blood transfusion, and presence of NFTI criteria. Models using SVI had the lowest AUC for all outcomes (Table 1). These differences persisted in adjusted models for ICU LOS ≥ 3 days, intubation, MVC, NAT, and sports (Table 2).

Conclusion: ADI and COI have higher predictive associations with pediatric trauma mechanisms and outcomes as compared to SVI. These data demonstrate a difference in model performance for predicting pediatric trauma mechanism and outcomes – with a higher AUC for models utilizing ADI and COI. This difference persists when adjusting for patient-specific metrics of socioeconomic disparity for multiple mechanisms and outcomes evaluated, suggesting neighborhood disparity is a driver of these mechanisms and outcomes. Future work is needed to validate these trends using national data.