150 - Validation of All Patient Refined-Diagnosis Related Group (APR-DRG) Risk of Mortality Score as a Severity Adjustor in Children

Poster Number: 150
Publication Number: 150.416

Balagangadhar R. Totapally, Nicklaus Children’s Hospital, Miami, FL, United States; Ramesh Sachdeva, Nicklaus Children’s Hospital, Miami, FL, United States

Background: There is an increase in the use of administrative databases for outcomes research in healthcare. The APR-DRG scores are proprietary risk adjustors calculated from discharge billing codes and are based on primary and secondary discharge diagnosis, age, and preexisting medical conditions, and are widely available in administrative databases. APR-DRG ranks the risk of mortality (ROM) and severity of illness (SOI) as low, medium, high, and extreme. Although there are some risk adjustment tools for critically ill patients there is a paucity of literature on the use of any method to adjust the risk of mortality across all pediatric discharges.

Objective: The objective of the study is to evaluate the performance of APR-DRG ROM as a risk adjuster for mortality in hospitalized children.

Design/Methods: We have used HCUP’s Kids’ Inpatient Database of 2016 for the study. We have excluded neonatal admissions and any cases with missing APR-DRG ROM variables. Binary regression analyses were performed with mortality as a dependent variable and ROM, SOI, and the number of complex chronic conditions (CCC) individually and in combination. For discriminant analyses, AUC of ROC was constructed with predicted mortality from regression analyses and observed mortality. Calibration was evaluated using the Brier score, average absolute error, and the means of expected and observed mortality.

Results: Out of a total of 6,266,285 weighted discharges during 2016, 2,294,728 were included in the analysis. The average age was 11 (7.4) years and 56.5% were females. The observed hospital mortality rate was 0.36%. The APR-DRG ROM showed excellent discrimination and calibration for observed mortality (Figure 1). The addition of SOI and the number of CCCs has not improved the model discrimination or calibration substantially (Table 1).Conclusion(s): APR-DRG ROM showed good discrimination and calibration for hospital mortality in children. It is a valid score for risk stratification in outcomes research using administrative databases.
Table 1. Discrimination and calibration data for APR-DRG ROM and SOI and CCC for validating hospital mortality prediction.
Figure 1. Receiver operating curves for APR-DRG ROM and SOI and CCC for predicting hospital mortality.