Identifying candidates for medical coding audits: demonstration of a data driven approach to improve medicare severity diagnosis-related group coding compliance

Abstract

Correct code assignment of Medicare Severity Diagnosis-Related Group (MS-DRG) is critical for healthcare. However, there is a gap currently on automatically identifying all potentially miscoded cases and prioritizing manual reviews over these cases. This paper reports a new process using a data-driven machine learning approach to flag potentially misclassified cases for manual review. We investigated using a stack of regularized logistic/softmax regression, random forest, and support vector machine to suggest potential cases for manual review by care providers, provided details addressing the data imbalance, and explored using features from various source including diagnosis and procedure codes, length of stay and the access log data from the electronic health record system. This potentially improves the efficiency of the coding review by care providers, providing another line of defense against miscoding to enhance coding compliance, and reduce the negative effects of upcoding and downcoding. We tested the new method with four common pediatric conditions and demonstrated its feasibility.