Data Mining Technology Helps Insurers Detect Health Care Fraud
Protecting financial assets from fraud or abuse is a universal and ongoing challenge facing the insurance industry. For those tasked with preventing, detecting, investigating and resolving these matters, data mining has become a key tool that maximizes both the effectiveness and efficiency of fraud investigations.
Data mining can be defined as the detection of relevant patterns, trends, relationships and correlations within a database. In the management of fraud and abuse issues, data mining techniques attempt to identify patterns within claims data that are different from patterns typically demonstrated by non-fraudulent individuals. Once the patterns have been identified, data analysis can take place. Data analysis is the process of analyzing mined data to interpret its significance.
Data mining can take many forms and range dramatically in complexity and sophistication. In its simplest use, data mining can involve simple mathematics (addition, subtraction, multiplication and division) and be applied to extracts of summary claim information. These simplistic calculations can be performed by investigators using basic desktop applications and are best utilized in the investigative phase, once the subject of an investigation has been identified.
Basic data mining can be used by investigators to screen and prioritize a case, or even prove or disprove an allegation.
Advanced Query Tools
In more advanced applications, high-end software packages can be used in conjunction with a companys data warehouse to perform ad hoc data extracts and analyze the subsequent claims payment information.
As part of ongoing business operations, most companies maintain a claims database and a query tool to pull information from the database. Often, these query tools can be adapted or modified to identify common indicators of fraud and abuse. Since these applications typically depend on a formal query or rule set, they are well suited to perform high-frequency, high-volume analysis in a standardized manner. In addition, they do not require capital investments beyond whats needed for ongoing business operations.
Unfortunately, this solution does have its drawbacks. These applications typically require an advanced understanding of data architecture and query language. Also, because the knowledge of specific schemes and their indicators is a pre-requisite for developing an effective query, developing a library of rules to anticipate every situation can be a time-consuming and costly process. Even if one did have the resources to build such a library, the process would essentially be unending, as new fraud schemes continually emerge and evolve.
Specialized Software Applications
Today there are numerous specialized software applications that focus exclusively on identifying suspected instances of fraud and abuse. In general, these applications can be used to analyze a specific universe or block of claims data. This block, often referred to as a “data set,” generally consists of historical claims information for a certain period of time and a certain peer group (i.e., the application might examine the claims submitted from every chiropractor in Kentucky for the last year).
Using a combination of advanced statistics specifically designed to identify indicators of fraud or abuse, the application will allow an analyst to sort rapidly through large amounts of information and identify patterns or entities that are statistically different from their peers and potentially indicative of fraud. Because these models are fairly fixed, this process is known as a “supervised” methodology.
While these applications have been used with moderate success, a chief drawback is that they are only effective on fraud schemes that already have been successfully perpetrated on the company. While you may have the opportunity to shut down an ongoing fraud scheme, it has been proven time and again that funds already lost to fraud are nearly impossible to recoup in their entirety. In order to fully mitigate significant losses, data mining and the detection of potential fraud and abuse needs to occur prior to funds being paid.
Next Generation Applications
Currently, several payers are experimenting with technology adapted from the credit card industry. This technology has several distinct advantages. First, the technology looks at historical claim submissions to build an ad hoc model to which the current claim can be compared. Next, the application compares the current claim to the historical model and assigns a relative score that reflects the claims similarity to the model. Then, that score is used to predict the potential of fraud. Since the application is not based on a set of static rules, it is considered an “unsupervised” methodology.
One of the great advantages of this new technology is its ability to be utilized prior to payment. This way, individual claims with a high likelihood of fraud can be forwarded to an investigator who can scrutinize the claim further and evaluate the appropriateness before payment is made. Investigations prior to payment are one of the most effective means of preventing fraud.
Ultimately, data mining tools and techniques cannot automate the entire process of managing fraud and abuse. However, what these tools can do is greatly increase the effectiveness of an organizations resources by enabling individuals to sift through mountains of data quickly and focus their efforts on the pertinent information most likely to generate savings.
is a director of Fraud and Abuse Operations for Ingenix Inc., Salt Lake City.
Reproduced from National Underwriter Edition, October 14, 2004. Copyright 2004 by The National Underwriter Company in the serial publication. All rights reserved.Copyright in this article as an independent work may be held by the author.