Detecting Medicaid Data Anomalies Using Data Mining Techniques
Shenjun Zhu, Qiling Shi, Aran Canes · 2010
The purpose of this study is to use statistical and data mining techniques in Base SAS(R) and SAS(R) Enterprise Miner TM to proactively reduce the number of false positives caused by data anomalies in Medicaid pharmacy claim data when employing a rule-based approach to identify overpayments. Typically rule-based techniques are based on specific state Medicaid laws and policies using certain formulas to detect and identify over charged payments. False positives are defined as an identified overpayment that is erroneously positive when a claim was paid correctly due to data anomalies or unknown factors. False positives substantially increase the amount of time and resources spent by the auditors. The specific objective of the study is to detect and reduce data anomalies by examining the relationships among key variables such as Medicaid amount paid (MAP), average wholesale price (AWP) and quantity of service in Medicaid pharmacy claim data. Pharmacy claim data were simulated and the overpayment was calculated by a rule-based approach developed by AdvanceMed Corporation. Different data mining techniques such as the studentized residual, leverage, Cook‟s distance, DFFITS and clustering were utilized to capture the abnormal claims and reduce the number of false positives. The results of this analysis indicated that the clustering statistical method is the best approach to detect these kinds of data anomalies, followed by the DFFITS method.