HSR 1: States, Medicaid, and the Structure of Health Care
580 - Classification of hospitals based on severity and scope of pediatric services
Saturday, April 29, 2023
3:30 PM – 6:00 PM ET
Poster Number: 580 Publication Number: 580.222
Troy Richardson, Children's Hospital Association, Lenexa, KS, United States; Jay Berry, Boston Children's Hospital, Harvard Medical School, Boston, MA, United States; Matt Hall, Children's Hospital Association, Lenexa, KS, United States; James C. Gay, Vanderbilt University Medical Center, Nashville, TN, United States
Vice President, Analytics & Data Strategy Children's Hospital Association Lenexa, Kansas, United States
Background: Hospitals are compared against industry benchmarks to assess their performance using measures of cost efficiency and quality of care. However, most administrative databases used in health services research do not contain a meaningful hospital classification based on the spectrum of services provided. Objective: Develop a meaningful classification of hospitals, incorporating the spectrum of services provided. Design/Methods: This was a retrospective cohort study of AHRQ’s national 2019 Kids’ Inpatient Database. After exclusion of normal newborns, obstetrical admissions, and hospitals with less than 100 admissions, we performed a k-means cluster analysis to group hospitals based on 1) neonatal case-mix index (N-CMI) and 2) non-neonatal hospital diagnostic diversity index (NN-HDDI), a normalized measure of number of non-neonatal conditions treated.We used these characteristics since 65% of hospital days in kids < 18 years are in newborns/neonates and prior research has shown HDDI is associated with hospitalization cost. Hospital characteristics were compared across clusters, and a linear discriminant analysis (LDA) estimated parameters from which any hospitals can be classified based on their observed N-CMI and NN-HDDI. Predictions from the LDA were compared against cluster results. Results: Using ~2.6 million hospitalized patients from neonates to age 20, 2,328 hospitals were clustered into 5 discrete hospital types. Cluster #1 was the smallest (118 hospitals, 1.3% of hospitalizations), provided no neonatal services, and cared for an older population of patients (median [IQR] age in years: 19 [15,19]).Cluster #2 included 1,156 hospitals (19.1% of hospitalizations) and provided a relatively limited diversity of diagnostic services.Cluster #3 included 674 hospitals (23.4% of hospitalizations) and had a similar diagnostic diversity compared with Cluster #2 but tended to care for a higher severity neonatal population indicated by a higher neonatal case-mix index.Cluster #4 included 238 hospitals (20.9% of hospitalizations) also cared for a higher severity neonatal population but had more than 2x the non-neonatal diagnostic diversity of the first 3 clusters.Cluster #5 included 142 hospitals, accounted for 35.8% of hospital admissions, and had the highest neonatal case-mix as well as the highest non-neonatal diagnostic diversity. The LDA model correctly classified >95% of hospitals into their respective clusters.
Conclusion(s): CMI and diagnostic diversity thresholds can be used to distinguish hospital types that can be used for comparing similar hospitals on cost and quality.
