Prescription drug abuse is one of the United States' most challenging fronts in combating illegal activity and protecting public health. The National Drug Intelligence Center estimated that recreational use of prescription drugs costs insurance companies as much as $72.5 billion each year, up to two-thirds of which is paid by public insurers. Catching culprits behind the trend is difficult to do without heaping false accusations on those with a legitimate need, but big data analytics are offering a new approach for controlling fraudulent behavior.
A recent Atlantic article reported that one prescription processing company is using analytics to catch fraudulent behavior that may be invisible on a patient level but that becomes clear with system-wide data. For instance, an Oxycontin abuser might visit 20 different doctors to obtain prescriptions, and each doctor would write the prescription in good faith. Similarly, a doctor might mask suspect prescriptions by varying the exact doses of drugs so that a simple review of data might not notice a pattern.
If a SQL database of prescription data were examined on a large scale with a Hadoop file system, however, these trends might become clear. The company has implemented this type of large scale analytics and has been able to uncover hubs of illegal activity.
Diving into unstructured data
One of the most valuable sources for knowing what doctors to examine more closely has been social media data, analysts told the Atlantic. People might share information on social networks about particularly lax doctors, and these posts can be "good indicators to have a suspicion," one analyst said. Being able to use data in any form is a cornerstone of today's analytics, particularly as tools such as Hadoop build information from text and other sources into reports.
According to insurance fraud technologist James Ruotolo, textual analysis is one of the best ways of spotting suspicious behavior, as many indicators that someone might be gaming the system come in the form of unstructured data. In an article for Insurance and Technology, he explained that although as much as 80 percent of insurance claim data is in unstructured, text-based formats, most predictive models only incorporate structured data.
"Most insurance companies are aware that they need better plans to deal with big data but many organizations still do not have a good handle on how to better leverage their growing repository of text-based information," Ruotolo wrote. "Experts agree that unstructured data is growing at an exponential rate and insurers are increasingly looking to use both internal sources, like claim notes, as well as external sources including social media."
With Hadoop big data analysis, unstructured data can be incorporated into attempts to understand fraudulent behavior or track activity such as illegal prescription use that can negatively impact insurers' bottom lines. Organizations that implement Hadoop can begin making the most of data at their disposal to curb unwanted behavior.