An unsupervised Bayesian hierarchical method for medical fraud assessment
Tahir Ekin, Greg Lakomski, R. Muzaffer Musal · Statistical Analysis and Data Mining The ASA Data Science Journal · 2019
The increasing size and complexity of health care industry makes it attractive for fraudsters, therefore medical fraud assessment has gained more importance. Statistical methods can help medical auditors reveal fraud patterns within medical claims data. This paper proposes an unsupervised Bayesian hierarchical method as a prescreening tool to aid in medical fraud assessment. The proposed hierarchical model helps the investigators group medical procedures and identifies the hidden patterns among providers and medical procedures. Outlier detection and similarity assessment are conducted to analyze the billing differences among providers. We illustrate the utilization of the proposed method using U.S. Medicare Part B data and discuss the potential insights for medical audit decision‐making.