Assessment on insurance fraud risk in basic medical insurance in the context of big data / 中国卫生政策研究

ABSTRACT

Objectives:To construct a fraud risk prediction model for basic medical insurance holders, discover the main characteristics of fraud,and then establish a risk assessment index system to provide decision support for an apposite supervision of medical insurance funds. Methods: Using the large-scale real data including more than 183 million records of basic medical insurance diagnosis and treatment in China,the integrated risk assessment model for basic medical insurance holders is constructed using XGBoost algorithm and EasyEnsemble method. On this basis, this paper further identifies and quantifies the potential characteristics of fraud enforcement, and thus constructs a fraud risk assessment index system. Results: The proposed integrated model predicted the fraud risk with the accuracy of 83％,balance predictive value of 95％,and the balance sensitivity was 85％,respectively. Most importantly, the probability of the insured fraud being correctly evaluated was 82％ in this fraud risk assessment model. Besides, the amount of various expenses incurred at each stage of assessment,and the number of various types of projects are important indicators to distinguish the fraud from the normal insurance holders. Conclusions: The fraud risk assessment index system constructed based on the XGBoost integrated model is effective for the identification of potential fraudsters among the basic medical insurance holders. Establishing a risk assessment index system and developing an apposite supervision system based on big data of medical insurance play an essential role in improving the level of medical insurance management services,which ensures the safety of medical insurance funds,and safeguards the social health insurance fairness.