TOP 24 - Investigating Predictors for Pediatric Emergency Department visit by Low Triage Patients (ESI i.e., Emergency Severity Index 4, 5)

Publication Number: 4727.TOP 24

Anika Naeem, University of Mississippi School of Medicine, Jackson, MS, United States; Cynthia Karlson, University of Mississippi Medical Center, Jackson, MS, United States; Christina Marbrey, Children's of Mississippi, Jackson, MS, United States; Fernando Gomez, University of Mississippi School of Medicine, Jackson, MS, United States; Md Abu Yusuf Ansari, University of Mississippi Medical Center, Jackson, MS, United States

Background: Approximately 30 million children visit emergency departments (EDs) in the U.S each year. Delivering efficient patient care while maintaining patient flow in the ED can be challenging. Many EDs are experiencing ever-increasing patient volumes, as they serve as a safety net for individuals without established care, resulting in ED overcrowding and a dilution of resources.

The Emergency Severity Index (ESI) system stratifies patients presenting to the ED on a scale from 1 to 5. Lower acuity patients (ESI Level 4 and 5) can often be safely seen in outpatient clinics. The use of ED resources by lower acuity patients is influenced by multiple factors, such as transportation issues, language barriers, and insurance status.

Objective: The aim of this study is to investigate the predictors of pediatric patients’ visits to state tertiary care hospital’s ED for low-acuity causes (ESI Level 4 and 5) by examining the time of presentation (clinic hours, 8 a.m.–5 p.m. vs. non-clinic hours, 5 p.m.–8 a.m.), day of visit (weekday vs. weekend), language preference (English vs. other), status of established care (primary care physician vs. none), and insurance status (Medicaid vs. other vs. none).

Design/Methods: This is a retrospective chart review cohort study (timeline: March 1, 2024 – March 1, 2025). Inclusion criteria include patients who presented to the pediatric ED aged 1 day to 18 years. Exclusion criteria include patients from the above-mentioned pool who were triaged as levels 4 or 5 and were admitted. Data was collected from electronic health records. The estimated sample size is approximately 50,000.

All analyses will be conducted at the encounter level. Exploratory data review will be performed using histograms, box plots, and density plots (for continuous variables) and frequency tables (for categorical variables).

Bivariate relationships will be explored using cross-tabulations, scatterplots, and correlation matrices to guide model building. For comparisons of continuous variables, a linear mixed-effects model with a random intercept for patients or a GEE with an identity link will be used. For comparisons of categorical variables, Rao–Scott–adjusted chi-square tests will be used.

The five primary hypotheses will be tested using multivariable GEE logistic regression with exchangeable correlation, estimating population-averaged odds ratios. As a sensitivity check, these models will be refitted as mixed-effects logistic regressions with a patient-level random intercept. All analyses will be performed in SAS 9.4 and R 4.2.2, and statistical significance will be set at α = 0.05.