Toward the Measure of Credibility of Hospital Administrative Datasets in the Context of DRG Classification

Abstract

Poor quality of coded clinical data in hospital administrative databases may negatively affect decision making, clinical and health care services research and billing. In this paper, we assessed the level of credibility of a nationwide Portuguese inpatient database concerning the codification of pneumonia, with a special emphasis on identifying suspicious cases of upcoding affecting proper APR-DRG (All-Patient Refined Diagnosis-Related Groups) classification and hospital funding. Using data on pneumonia-related hospitalizations from 2015, we compared six hospitals with similar complexity regarding the frequency of all pneumonia-related diagnosis codes in order to identify codes that were significantly overreported in a given facility relatively to its peers. To verify whether the discrepant codes could be related to upcoding, we built Support Vector Machine (SVM) models to simulate the APR-DRG system and assess its response to each discrepant code. Findings demonstrate that hospitals significantly differed in coding six pneumonia conditions, with five of them playing a major role in increasing APR-DRG complexity, being thus suspicious cases of upcoding. However, those comprised a minority of cases and the overall credibility concerning upcoding of pneumonia was above 99% for all evaluated hospitals. Our findings can not only be relevant for planning future audit processes by signalizing errors impacting APR-DRG classification, but also for discussing credibility of administrative data, keeping in mind their impact on hospital financing. Hence, the main contribution of this paper is a reproducible method that can be employed to monitor the credibility and to promote data quality management in administrative databases.