دو مدل برای بررسی کلاهبرداری مدیکر در داخل پایگاه داده های بدون نظارت

We propose two models to identify fraud, waste and abuse in Medicare. These models are used to flag health care providers. The motivation for these models is based on observed cases of fraud. The paper details the use of clustering algorithms, regression analysis, and various descriptive statistics that are components of these models. Some of the challenges in the struggle to reduce fraud in Medicare are discussed.

It is clear that fraud plays an important role in the US healthcare system. It increases the cost of healthcare through direct and indirect means. Its direct impact includes fraudulent monetary charges to the healthcare system. It also has indirect impacts which mainly arise from false positive identifications of fraudulent health care providers. These include opportunity costs associated with the medical education of the fraudulent providers and costs associated with the construction of complicated policies that effect beneficiaries and providers alike. The general goal of organizations charged with fighting fraud can be formulated as reducing direct and indirect costs by maximizing the percentage of correct identification of fraudulent providers while minimizing the false ones with the least amount of resources. This paper exposes the application of two methodologies to identify fraudulent infusion therapy drug providers in a number of states from an unsupervised database. The models employed in fraud identification can be generalized into two categories. There are models that identify fraud after observing nearly irrefutable amount of evidence, for instance identifying providers that charge for services rendered amounting to more than 24 h a day. On the other hand there are models which can identify providers of interest as being possibly involved in fraud after observing abnormal patterns in the data. The false positive rates for these models are likely to be higher. The models exposed in this paper belong to the latter category where we argue for their necessity in decreasing the overall level of fraud in the healthcare system. The paper is organized in six sections. Section 2 is an overview of healthcare fraud investigations within US and provides a review of recent work in the literature but is not meant to be a comprehensive literature review. Section 3 provides an overview of data and Section 4 provides the exposition of the two models that we use to investigate fraud. In Section 5 we demonstrate the application of the methods. Section 6 includes the results and conclusion.

In this paper we listed two methods to identify fraud in Medicare infusion therapy providers. We do not make the claim that every health care provider flagged in this study is involved in fraud. On the other hand, we argue that these methodologies which identify possible cases of fraud are useful as the first line of defense mechanism in identifying fraud. We believe a system dynamic approach is required to investigate the Medicare system in decisions involving the investigation of possible providers of fraud. We suggest that a provider who gets flagged as a result of an unsupervised fraud detection method needs to go through a second stage that should cause him/her the least disutility.