Stritch Professor Uses AI to Identify Patients with Unhealthy Substance Use

Loyola University Chicago Stritch School of Medicine Assistant Professor Majid Afshar, M.D., cares for critically ill patients, many with life-threatening infections and organ failures. But what he and other health care professionals often don’t immediately address is a patient’s unhealthy use of alcohol, opioids or prescription medications.

“When a patient comes in with a pneumonia and mentions he drinks several beers a day, the information is usually documented in a patient’s electronic health record.  But the focus is on treating the pneumonia, not the unhealthy use of alcohol,” explains Dr. Majid.  

Unhealthy substance use is a major public health problem in Chicago, straining resources at hospitals and at the city.  Since 2011, Chicago has led the United States in opioid-related Emergency Department visits. Between 2015 and 2016, the city’s opioid-related deaths jumped by 74 percent. Unhealthy alcohol use reported by patients is as high as 30 percent.

Dr. Afshar explains that patients with undetected, unhealthy substance are more prone to complications during hospitalization.  Complications generally translate into longer hospitals stays, higher readmission rates, and increased costs.  According to AHRQ’s Healthcare Utilization Project, alcohol-related diagnosis is the second leading cause of seven-day hospital readmissions.

Concerned about the impact of unhealthy substance use, the Chicago Department of Public Health developed the Healthy Chicago 2.0 Plan that seeks to reduce opioid-related overdoses and adult binge drinking rates. The plan includes leveraging health care technology through partnerships like the Institute of Translational Medicine, which includes Loyola University Chicago and Dr. Afshar, who is also a clinical informatics researcher.

Dr. Afshar believes artificial intelligence (AI) can plan an important role in meeting the city’s goals. With AI, computers are “taught” using machine learning to scan patients’ electronic health records (EHRs) for words and data captured during routine care that identify people at risk of alcohol or opioid-related complications. If a patient’s unhealthy substance use is flagged during hospitalization, targeted interventions may be taken and complications and subsequent costs may be avoided.

As easy as this sounds, Dr. Afshar says the process is still being perfected. Researchers only have access to about 25 percent of the data in EHRs, data typically in a checkbox format. The balance of the information is in the form of typed notes unique to the physician and therefore not easily summarized. Artificial intelligence such as natural language processing enable computers to “read” the text in EHRs, accurately identify individuals who misuse alcohol or opioids, and notify the care teams to act.

Dr. Afshar and a multi-disciplinary team of experts from across Loyola University’s Health Sciences Campus have developed the AI tool.  Early indications are promising. The team has validated misuse classifiers for alcohol and opioids with a sensitivity and specificity above 70 percent.

To test their results and support future real-time implementation, the team submitted a grant application to the Center for Health Outcomes and Informatics Research at Loyola University Chicago and is working on a second grant from the National Institutes of Health that will leverage the Clinical Research Database at Loyola University Medical Center and incorporate additional data domains from EHRs to differentiate levels of risk from unhealthy alcohol and opioid use to better guide physicians and other members of a health care team.  

Using AI, immediate physical needs will be treated and unhealthy substance use and the potential for complications also may be addressed. Eventually, AI may be used to identify other illicit drug use.  

While Dr. Afshar’s research is focused on Loyola Medical Center’s data, he plans to expand the research, and its potential benefits, to partners within the Institute of Translational Medicine (ITM), including the University of Chicago, Rush University, North Shore Health System, and Advocate Health Care. Thanks to the ITM partnership, open-source tools and high performance computing, they now can build interoperable information technology systems to support AI research and implement solutions that may benefit many Chicagoans.

“The data we gather will enable us to see patterns of unhealthy substance use across health systems that serve Chicago—or anywhere else for that matter—which will help public health agencies and health systems to direct services and solutions to very specific areas and facilities,” says Dr. Afshar.

Identifying and addressing unhealthy substance use at the point of care is important.  “We want to make sure every person who is hospitalized receives the right treatment and has access to behavioral health treatment services once they leave,” he says.