Development of a pediatric readmissions encounter predictor: Benchmarks for 30-day unplanned pediatric readmission

Abstract

Pediatric readmissions are important for hospitals to measure and monitor to identify potential improvement opportunities but require benchmarks to contextualize observed and expected readmissions. We developed and validated the Pediatric Readmissions Encounter Predictor (PREP) in administrative data as pediatric readmission benchmarks using the All-Patient Refined Diagnosis-Related Groups with severity of illness subclasses. We developed the model using data from the National Readmission Database in 2019. We subsequently validated these models in the same data set from 2018. The overall 30-day unplanned readmission rate was 2.5%. The model demonstrated acceptable discriminatory performance (area under the receiver operator characteristic curve [AUC] = 0.738, 95% confidence interval [CI]: 0.734, 0.741) and was well calibrated across all levels of predicted probabilities. PREP is a promising readmission prediction model which hospitals can use to assist in identifying improvement opportunities.