A prescription fraud detection model

摘要

Prescription fraud is a main problem that causes substantial monetary loss in health care systems. We aimed to develop a model for detecting cases of prescription fraud and test it on real world data from a large multi-center medical prescription database. Conventionally, prescription fraud detection is conducted on random samples by human experts. However, the samples might be misleading and manual detection is costly. We propose a novel distance based on data-mining approach for assessing the fraudulent risk of prescriptions regarding cross-features. Final tests have been conducted on adult cardiac surgery database. The results obtained from experiments reveal that the proposed model works considerably well with a true positive rate of 77.4% and a false positive rate of 6% for the fraudulent medical prescriptions. The proposed model has the potential advantages including on-line risk prediction for prescription fraud, off-line analysis of high-risk prescriptions by human experts, and self-learning ability by regular updates of the integrative data sets. We conclude that incorporating such a system in health authorities, social security agencies and insurance companies would improve efficiency of internal review to ensure compliance with the law, and radically decrease human-expert auditing costs.