Network Analysis to Define Pediatric Acute Care Regions in Wisconsin.

Abstract

Objective: To pilot a system for deriving borders of pediatric regions, and to compare these to adult markets based on fit with pediatric utilization data.

Study setting and design: In this cross-sectional study, we studied all acute care encounters (emergency department visits and hospitalizations) for children less than 16 years old in Wisconsin 2021-2022.

Data sources and analytic sample: We used the Healthcare Cost and Utilization Project State Emergency Department and Inpatient Databases. We first counted how many patients from each ZIP code visited each hospital and mapped ZIP-hospital connections. Using a network analysis technique called community detection that clustered hospitals by their common connections, we grouped ZIP codes to form pediatric emergency service areas (PESAs). We counted patient referrals within and between PESAs and repeated the community detection procedure, resulting in pediatric emergency referral regions (PERRs). The primary outcome was modularity, a common network fit measure ranging from -1 to 1 (1 represents perfect clustering). We also compared demographics and network quality measures between PERRs, hospital referral regions (HRRs), core-based statistical areas, and Pittsburgh Trauma Atlas regions.

Principal findings: We analyzed 587,886 encounters, from which ZIP codes grouped into 24 PESAs. Based on referral patterns, there were 4 PERRs. PERRs had modestly higher modularity for interhospital referral patterns than all other systems (0.53, 95% confidence interval [CI] 0.52, 0.54 compared to 0.46, 95% CI 0.46, 0.47 for HRRs). PERRs were larger (median 11,361 mile² vs. 3957 for HRRs), contained more children (median 265,222 vs. 49,667 for HRRs), and contained more hospitals (median 35 vs. 7 for HRRs) than all other systems.

Conclusions: Using Wisconsin HCUP data, we derived pediatric acute care regions with a strong fit for pediatric utilization data. Future work should test this approach across the whole US, which would allow between-region cost and outcomes comparison.