Predicting Low-Resource-Intensity Emergency Department Visits in Children

Abstract

Objectives: Interventions to reduce frequent emergency department (ED) use in children are often limited by the inability to predict future risk. We sought to develop a population-based model for predicting Medicaid-insured children at risk for high frequency (HF) of low-resource-intensity (LRI) ED visits.

Methods: We conducted a retrospective cohort analysis of Medicaid-insured children (aged 1-18 years) included in the MarketScan Medicaid database with ≥1 ED visit in 2013. LRI visits were defined as ED encounters with no laboratory testing, imaging, procedures, or hospitalization; and HF as ≥3 LRI ED visits within 365 days of the initial encounter. A generalized linear regression model was derived and validated using a split-sample approach. Validity testing was conducted examining model performance using 3 alternative definitions of LRI.

Results: Among 743,016 children with ≥1 ED visit in 2013, 5% experienced high-frequency LRI ED use, accounting for 21% of all LRI visits. Prior LRI ED use (2 visits: adjusted odds ratio = 3.5; 95% confidence interval, 3.3, 3.7; and ≥3 visits: adjusted odds ratio = 7.7; 95% confidence interval, 7.3, 8.1) and presence of ≥3 chronic conditions (adjusted odds ratio = 1.7; 95% confidence interval, 1.6, 1.8) were strongly associated with future HF-LRI ED use. A model incorporating patient characteristics and prior ED use predicted future HF-LRI ED utilization with an area under the curve of 0.74.

Conclusions: Demographic characteristics and patterns of prior ED use can predict future risk of HF-LRI ED use in the following year. Interventions for reducing low-value ED use in these high-risk children should be considered.

Keywords: emergency medicine; pediatrics; predictive model; utilization.