A first-of-its kind study has demonstrated that an artificial intelligence technique can be used to identify trauma patients who misuse alcohol.
Researchers from Loyola Medicine and Loyola University Chicago used the technique, natural language processing, to identify alcohol misusers from clinician notes in electronic health records.
In 78 percent of cases, the technique was able to differentiate between patients who misused alcohol and those who did not. Corresponding author Majid Afshar, MD, MSCR, and colleagues published their findings in the Journal of the American Medical Informatics Association.
Dr. Afshar, a Loyola Medicine critical care physician, is an assistant professor in the Division of Pulmonary and Critical Care Medicine and Department of Public Health Sciences of Loyola University Chicago Stritch School of Medicine.
The study was a cross-campus collaboration that included researchers from Loyola's Burn and Shock Trauma Research Institute, Center for Health Outcomes and Informatics Research, Department of Public Health Sciences, Department of Computer Science, Department of Medicine and Department of Surgery.
As many as 1 in 3 trauma patients misuse alcohol, and many trauma cases are alcohol related. Previous research has shown that a traumatic injury provides an opportunity for a teachable moment. Screening, brief intervention and referral to treatment (SBIRT) can reduce subsequent alcohol consumption, decrease injury recurrence by nearly 50 percent and reduce rates of DUI arrests.
The brief intervention typically includes providing information on the link between drinking and injury, encouraging patients to think about how drinking may have contributed to their injuries and giving professional advice about the need to reduce risk by cutting down or quitting drinking.
Current screening methods employ the 10-item Alcohol Use Disorders Identification Test (AUDIT). But there are drawbacks to this screening test. Patients may not be honest when answering questions about their alcohol use or may not be able to communicate at all. Staff may not be available to administer the test, especially during nights and weekends. In addition, screening is a "resource-intensive process that imposes significant costs on a health system," Dr. Afshar and colleagues wrote.
Using artificial intelligence to screen for alcohol misusers potentially could overcome these problems. To test this idea, researchers sifted through electronic health records using natural language processing and machine learning. The artificial intelligence technique employs computational methods that help computers understand human language.
The study included records of 1,422 adult patients admitted to Loyola's Level 1 trauma center over 3 ½ years. The data included 91,045 clinician notes in electronic health records. The notes contained 16,091 medical concepts. Using natural language processing, researchers identified 16 medical concepts that predict for alcohol misuse.
The concepts include intoxication, neglect, drinking problems, liver imaging, sexually active, marijuana and the B1 vitamin thiamine. (Hospital patients with alcohol dependence commonly are treated for thiamine deficiency.)
The artificial intelligence technique likely would be affordable to trauma centers that have the expertise to use it, Dr. Afshar said. He noted that the open-source programming and linguistics software used by researchers would be free to any user.
Natural language processing "has adequate predictive validity for identifying alcohol misuse in the trauma setting," Dr. Afshar and colleagues concluded. The technique "provides an automated approach to potentially overcome staffing and patient barriers for SBIRT programs at trauma centers."
This article has been republished from materials provided by Loyola University Medical Center. Note: material may have been edited for length and content. For further information, please contact the cited source.
Reference: Afshar, M., Phillips, A., Karnik, N., Mueller, J., To, D., Gonzalez, R., … Dligach, D. (2019). Natural language processing and machine learning to identify alcohol misuse from the electronic health record in trauma patients: development and internal validation. Journal of the American Medical Informatics Association, 26(3), 254–261. https://doi.org/10.1093/jamia/ocy166