Variation in Utilization of Intensive Care for Pediatric Diabetic Ketoacidosis

Objective: To examine the hospital-level variation in intensive care unit (ICU) utilization and quantify the relative contribution of patient and hospital characteristics versus individual hospital factors to the variation in ICU admission rates among pediatric hospitalizations with diabetic ketoacidosis (DKA). Methods: The Texas Inpatient Public Use Data File was used to identify hospitalizations of state residents aged 1 month to 19 years with a primary diagnosis of DKA between 2005 and 2014. Multilevel, mixed-effects logistic regression modeling was performed to examine the association of patient- and hospital-level factors with ICU admission. Risk and reliability adjustment was then performed to assess hospital-level variation in ICU utilization. Intraclass correlation coefficient was used to quantify variation in use of ICU attributable to individual hospitals. The association between adjusted rates of ICU admission and total hospital charges and length of stay was examined using linear regression. Results: Of the 23 585 DKA hospitalizations, 14 638 (62.1%) were admitted to ICU. On multilevel analysis, the odds of ICU admission progressively decreased with rising volume of DKA hospitalizations (adjusted odds ratio: 0.08 [highest vs lowest quartile]; 95% confidence interval [CI]: 0.03-0.24). The crude median (interquartile range [IQR]; range) of ICU admissions across hospitals was 82.6% (73%-90%; 11.1%-100%). The median (IQR) risk- and reliability-adjusted ICU admission rate was 81.0% (73.0%-86.9%), ranging from 11.2% to 94%. Following risk and reliability adjustment, the intraclass correlation coefficient was 0.005 (95% CI: 0.004-0.006). For each 10% increase in adjusted ICU admission rate, total hospital charges rose by 7% (95% CI: 3%-11%). There was no association between ICU admission rates and hospital length of stay. Conclusion: Although high variation in ICU utilization was noted across hospitals among pediatric DKA hospitalizations, the proportion of variation attributable to individual hospitals was negligible, once adjusted for patient mix and hospital characteristics.