Carnegie Mellon University

Image of a pill bottle

November 29, 2018

Disrupting Opioid Addiction

Students Aim To Create Preemptive Tool To ID High-risk Users

By Lauren Prastien

Lauren Prastien
  • Heinz College of Information Systems and Public Policy

Carnegie Mellon University students have created a new tool to help clinicians identify individuals with high-risk patterns of opioid use before it's too late.

Using prescription data provided by the Allegheny County Department of Human Services (DHS), public policy and management graduate students Wilson Mui and Nikita Setia and Riccardo Fogliato, a Ph.D. candidate in the Department of Statistics and Data Science, created a predictive model that could project an individual's trajectory after just a few months of observation. This tool has the potential to save lives, especially in counties such as Allegheny, where the rate of opioid-related overdoses exceeds both the state and national averages.

"Our goal is to use data to try to find people before they have a problem," Mui said.

The students worked under the guidance of Daniel Nagin, a professor at CMU's Heinz College of Information Systems and Public Policy, and Dr. Jonathan Elmer, an assistant professor of emergency medicine, critical care medicine and neurology at the University of Pittsburgh. The project was funded by the Deloitte Foundation.

For Erin Dalton, deputy director for the Office of Data Analysis, Research and Evaluation at DHS, being able to locate high-risk opioid users before they develop a problematic dependence is absolutely critical.

"While other strategies allow us to identify people who are already addicted or actively in need of treatment, the fact that this method is preemptive is particularly valuable," said Dalton, an alumna of Heinz College.

The students synthesized eight years of DHS data into three distinct opioid user profiles: low users, heavy users and "desisters," or individuals whose prescription opioid usage starts high but then quickly diminishes over time. These profiles were based on the number and dosage of prescriptions filled on a monthly basis, as well as the changes in this data over time.

"Addiction is a disease, and like any other disease, prevention is always better than treatment," said Elmer, who has witnessed the devastating impact of opioid addiction in Allegheny County.

"Today, clinicians and families both are often shocked by unanticipated opioid overdoses and death," Elmer said. "Although the data we analyzed in this work are available to clinicians, advanced modeling techniques can identify patterns inapparent to the naked eye. By predicting future patterns of opioid use, our hope is to bring clinicians the tools they need to intervene early, modify high risk behaviors and save lives."

According to the U.S. Department of Health and Human Services, more than 11.4 million people in the U.S. misused prescription opioids in 2016 alone. The White House Council of Economic Advisors found that 2.4 million Americans currently have an opioid use disorder, costing the United States over $504 billion each year.

Nagin, whose group-based trajectory modeling algorithm served as the foundation for the team's opioid user profiling methodology, maintains this technique epitomizes the mission of Heinz College.

"This isn't the sort of problem that can be accurately captured from purely a policy perspective or an informatics perspective. Its complexity necessitates Heinz's interdisciplinary approach," Nagin said.

With better than 80 percent accuracy after just a few months of prescription opioid use, the students' model could help clinicians on the frontlines of the opioid crisis by flagging many high-risk individuals before they develop an opioid use disorder.

"We're only looking at their prescription data over time. We're not looking at things like gender or race. We are only looking at prescription use to establish a pattern," Setia said. "The model not only gives you the likelihood that a person is going to fall into one of the three groups for any given month, it gives you a picture of where the person is likely to fall in the long term."

By catching problematic behavior early, this method could facilitate gentle interventions from clinicians, such as conversations with high-risk patients and prescription strength adjustments, before the issue escalates, Elmer said.

One particularly promising aspect of this work is that it could scale nationally. The DHS dataset at the heart of this study is Medicaid data, meaning that every state has access to the same information. While there is no substitute for a clinician's presence and discretion, this work provides a more standardized benchmark for high-risk opioid usage.

"This isn't the end-all, be-all of the drug problem. It's not a one-stop solution," Setia said. "But there's bias that enters into clinicians' decision-making. One clinician's definition of 'heavy usage' may not match another's. This approach takes that out of the picture, giving standard definitions and letting the clinician handle it from there."

Currently, the majority of interventions into opioid misuse take place after the patient has developed an addiction. By targeting opioid misuse at such an early stage, this groundbreaking model would help clinicians take a preemptive, intervention-based approach to opioid addiction.