Using Data to Reduce and Predict Prescription Drug Fraud

July 2017
Topics: Health IT, Data Management, Information Management, Public Health (General), Law Enforcement, Clinical Medicine
Our researchers have developed a powerful analytical tool to predict and identify prescription drug fraud schemes involving patients, prescribers, and pharmacies. The system shows promise for state and federal government efforts to combat opioid abuse.
Prescription drugs on a table

In the United States, there was a 200 percent increase in opioid-related deaths from 2000 to 2014. In 2014 alone, more than 46,000 people died from a drug overdose, 28,000 of those stemming from opioid abuse. The National Health Care Anti-Fraud Association estimates the monetary cost to insurers, Medicare, and Medicaid—just from drug diversion and illicit acquisition—to be upward of $72 billion per year.

Both federal and state government agencies are creating programs to address this problem. One of these is the Prescription Drug Monitoring Program (PDMP), through which states are collecting data about transactions of controlled substances.

MITRE researchers saw the PDMP as a potential window into the problem of reducing drug abuse. Was it possible to employ advanced analytics on this data and develop a capability to help identify prescription fraud schemes involving patients, prescribers, and pharmacies? The state of Indiana agreed to partner with MITRE to find out.

The MITRE team, led by Jaya Tripathi, applied several analytic methods to 18 months of PDMP data from Indiana. The team developed techniques to identify individuals with suspicious drug-seeking behavior and "doctor shoppers" who re-sell the drugs on the streets, as well as physicians and pharmacies who aid and abet such activities.

"Employing the results from the techniques applied to the data in our research, we have created a web-based tool to help organizations analyze this kind of data to identify suspicious behavior," Tripathi says. "We want to get in front of the problem by predicting where fraud is likely to occur. We call it the Fraud Investigator's Analytic Tool [FIAT].

"We've deployed the tool at a state law enforcement agency in Indiana, and we continue to add features to make it more powerful. FIAT runs on a data model that has basic prescription data elements and could be useful to many organizations in government, insurance, law enforcement, claims auditing, community outreach, and hospital compliance networks."

Pinpointing the "Holy Trinity" of Prescription-Drug Abuse

The MITRE team focused on fraud schemes involving controlled drugs. "We specifically looked at instances in which three drugs [an opioid, a benzodiazapine, and a muscle relaxant] were prescribed at the same time to an individual," Tripathi explains. "This combination is known in the drug enforcement community as the 'holy trinity.'

"We also looked at other combinations such as 'speedballs' or 'triple threats,' which would never be prescribed for a medical reason, as they pose a heightened chance of respiratory depression and can lead to coma or death."

The states' collection of PDMP data has been a big help in looking at the scope of the problem. These data sets contain information about patients, prescribers, pharmacies, drugs, dosage, dates, times, and addresses.

At the time of the acquisition of the data from Indiana PDMP, the retail pharmacies in the state (as well as mail order pharmacies shipping to Indiana residents) were required to report any prescriptions filled with controlled substances at least every three days, including transactions involving private and public insurers and cash. The law has since changed so that now they have to report this information every 24 hours.

Highlighting Hot Spots of Potential Fraud

The MITRE team looked at 12 million transactions from October 2011 through March 2013, which involved 68,000 prescribers, 2,000,000 patients, and 1,200 pharmacies. This research project drew on MITRE's expertise in health IT, data analysis, coding, and data visualization. Team members tried out many different methods for analyzing data to find the best approaches. They also talked to experts in the field—including doctors, pharmacists, insurers, and law enforcement officials—to establish "ground truth." This helped them understand what their analyses meant and if they were accurately portraying the big picture.

In the data exploration stage, the team did statistical analysis, pulled the data into histogram charts and scatter plots, and employed clustering as well as some advanced visualization techniques.

Red flags quickly popped up. For example, one pharmacy had filled more than 1,000 holy trinity prescriptions for the same individual. And many patients were traveling up to 100 miles to go to one pharmacy in a different state.

Our researchers also identified a doctor who, according to the data, had prescribed the highest number of holy trinity drug cocktail prescriptions in the state.

"After we identified this prescriber using graph analytic techniques, we later found out that the doctor had been indicted—and that 31 of his patients died of overdoses," Tripathi says. "This demonstrated that our findings correlated with actual events, which validated our approach."

The team also created geospatial maps that showed hot spots of activity, for individual drugs as well as drug cocktails. For example, "the area identified as a hot spot on the map for oxymorphone later experienced an HIV epidemic," she adds. "The cause was attributed to the sharing of needles for injecting oxymorphone." Another hot spot showed 270,000 instances of holy trinity cocktails issued in one small area that falls between college campuses.

The team has recently acquired data from January 2014 to December 2016 and will be doing further research including a comparative study on trends.

Sharing Our Tools on a National Stage

"We feel FIAT, along with the graphical and geospatial techniques, can easily be used by many organizations that are fighting drug abuse and prescription fraud," Tripathi says. "For example, law enforcement agencies could use FIAT to detect fraudulent activity or high-risk behavior. The geospatial maps can be used by states to create public policies and to better plan where to put resources, such as community education programs or diversion officers. Preventing fraud could save both costs and lives."

The MITRE team has also produced two other tools:

  • A tool for decision-making designed to help prescribers and pharmacies discern risky behavior, such as doctor shopping. This is being tested by Brigham & Women's Hospital in Boston and the University of Maryland Medical School.
  • Geospatial Technique Tools, which help users see where the problem areas are geographically so that they can place more resources in those areas.

"We presented the results from our research to the White House Office of National Drug Control Policy and to the Office of Management and Budget at the White House," Tripathi says. "The Indiana Health Commissioner, Dr. Jerome Adams, remarked that the geospatial tools to produce heat maps that light up areas of concern would be a useful tool employed by all states, especially those organizations that struggle to allocate limited resources in the best way possible."

The team continues to demonstrate its tools to interested organizations, looking for those that want to collaborate on its further development.

—by Beverly J. Wood

To learn more or to collaborate with MITRE's FIAT team, contact us:


Publication Search