- Researchers used natural language processing and machine learning methods to recognize alcohol misusers from clinician notes in electronic health records.
- In 78% of cases, AI was able to distinguish between trauma patients who misused alcohol and those who didn’t.
As many as 10% of deaths in the United States happens due to alcohol misuse, and this rate of misuse increased by 9% between 2002 and 2012. In fact, 1 in 3 patients with trauma encounters has alcohol misuse.
Previous studies have proven that every traumatic injury provides an opportunity for a teachable moment. SBIRT (short for Screening, Brief Intervention, and Referral to Treatment) programs at trauma centers have been shown to decrease alcohol consumption and reduce injury recurrence by almost 50%.
Now, researchers at Loyola University Health System have demonstrated that artificial intelligence (AI) can be used to detect trauma patients with alcohol misuse. They used natural language processing (NLP) and machine learning methods to recognize alcohol misusers from clinician notes in electronic health records.
How It’s Different From Current Screening Methods?
The existing screening methods employ the 10-item AUDIT (short for Alcohol Use Disorders Identification Test), however, there are several drawbacks of these methods. It requires creating new forms and procedures into electronic health record systems.
Sometimes patients are not very honest when answering questions related to their alcohol use, thus test results might not be completely genuine. In addition, hiring employees to implement and administer the tools is a resource-intensive and time-consuming process.
NLP and machine learning algorithms have been successfully used in clinical practice and research. More specifically, NLP techniques rely on supervised learning that utilizes current reference standards to forecast unseen cases.
Reference: American Medical Informatics Association Journal | doi:10.1093/jamia/ocy166 | Loyola University Health System
Along with the providers’ documentation, the NPL classifier leverages proxy reports, laboratory data, embedded medication, and additional notes from other medical staff that are collected within the first 24 hours.
In this study, researchers used the data of 1,422 adult patients admitted to Loyola trauma center over three and a half years. The data contained more than 91,000 notes in electronic health records, which included over 16,000 medical concepts.
They used the Knowledge Extraction System and clinical Text Analysis to perform linguistic processing of notes. This helped them identify 16 medical concepts — including B1 vitamin thiamine, marijuana, liver imaging, and marijuana — that indicate alcohol misuse. In 78% of cases, AI was able to distinguish between trauma patients who misused alcohol and those who didn’t.
Overall, this AI provides an automated process to overcome patient and staffing barriers for SBIRT programs at health centers. It would be affordable to trauma centers that have the expertise to use this technique.
Researchers also mentioned that the linguistics software and open-source programming used in this study would be free to all users.