George Mason University

News at Mason

New method could help health care providers predict unreported opioid abuse

June 7, 2018   /   by Jiaxi Zhang

Mason's Farrokh Alemi and Sanja Avramovic and a colleague from NYU analyzed data from electronic health records to uncover correlations between patients’ diseases and substance abuse. Creative Services photos.

A new study published in Big Data by George Mason University researchers Farrokh Alemi and Sanja Avramovic could help clinicians predict unreported opioid abuse by their patients.

Alemi and Avramovic in Mason’s College of Health and Human Services and Mark Schwartz at New York University analyzed data from electronic health records to uncover correlations between patients’ diseases and substance abuse.

The researchers found that a variety of mental illnesses can be caused by or lead to substance abuse. For example, depression increased odds of substance abuse. 

Other indicators of potential substance abuse were certain physical illnesses, drug seeking behavior, self-injury and a weakened immune system. For instance, a patient who repeatedly gets common infections or has difficulty recovering from them is likely to be experiencing the consequences of substance abuse.

“What we’re hoping to do is start a conversation between health care providers and patients that can lead to treatment,” Alemi said. “By focusing on the medical complications from substance abuse, we can predict risk of future use and find it earlier than if we were following behaviors such as ‘doctor shopping.’ This approach could also help reduce the stigma around substance abuse and help the patients see the connection between their substance abuse and the negative health conditions they are experiencing.”

Data from 4.8 million veterans from Veterans Affairs Informatics and Computing Infrastructure from 2006 through 2016 were used to predict the patients’ diagnoses and the change in odds of substance abuse. The average age for the study cohort was 59.45 years. The majority of patients were male (96.5 percent) and white (34.5 percent).

“Automatically explaining the results of the screening remains our goal,” Avramovic said. “While we have yet to put this in clinical practice and explain our predictions, we have established that the proposed method is practical and accurate.”

In the future, the researchers are interested in examining whether computers could be used to communicate the screening results to both clinicians and patients.

Jiaxi Zhang is a graduate student working in the College of Health and Human Services.