Measuring data credibility and medical coding: a case study using a nationwide Portuguese inpatient database

Some countries have adopted the diagnosis-related groups (DRG) system to pay hospitals according to the number and complexity of patients they treat. Translating diseases and procedures into medical codes based on international standards such as ICD-9-CM or ICD-10-CM/PCS is at the core of the DRG systems. However, certain types of coding errors undermine this system, namely, upcoding, in which data is manipulated by deliberately using medical codes that increase patient's complexity, resulting in higher reimbursements. In this sense, ensuring data credibility in the context of upcoding is critical for an effectively functioning DRG system. We developed a method to measure data credibility in the context of upcoding through a case study using data on pneumonia-related hospitalizations from six public hospitals in Portugal. Frequencies of codes representing pneumonia-related diagnosis and comorbidities were compared between hospitals and support vector machine models to predict DRGs were employed to verify whether codes with discrepant frequencies were related to upcoding. Data were considered not credible if codes with discrepant frequencies were responsible for increasing DRG complexity. Six pneumonia-related diagnoses and fifteen comorbidities presented a higher-than-expected frequency in at least one hospital and a link between increased DRG complexity, and these targeted codes was found. However, overall credibility was very high for nearly all conditions, except for renal disease, which presented the highest percentage of potential upcoding. The main contribution of this paper is a generic and reproducible method that can be employed to monitor data credibility in the context of upcoding in DRG databases.