I've written before about some interesting applications of big data and AI to predict which patients might progress to substance abuse, with a team from the University of Colorado developing technology to predict potential opioid abuse among patients.

A second project, from Loyola Medicine and Loyola University Chicago, now aims to do a similar thing for alcohol abuse. The work, which was documented in a recently published paper, uses natural language processing to assess each patient's electronic medical records to identify potential alcohol misusers.

The authors reveal that their tool was able to accurately differentiate patients who had misused alcohol and those who had not in around 78 percent of cases.

Misuse of Alcohol

The authors reveal that around 1 in 3 trauma patients misuse alcohol in some way, but that the injuries that result from the trauma represent a teachable opportunity. Evidence suggests that referring trauma patients to brief interventions can reduce recurrence rates by up to 50 percent.

This intervention typically involves providing the patient with clear information on the connection between alcohol and injury, whilst encouraging them to think about how their drinking has contributed to their injuries, with professional advice then given to help cut down or quit.

At the moment, screening typically involves something like the Alcohol Use Disorders Identification Test (AUDIT), but the test only really works if the patients are honest when answering the questions. Indeed, many patients are in no fit state to answer at all. The team believes that AI can do a better job.

They tested their system by feeding it nearly 1,500 electronic health records so that it could begin to spot patterns that indicate alcohol abuse. In total, the records had over 90,000 clinician notes in them, with the system able to focus in on 16 medical concepts related to alcohol misuse, including liver imaging and neglect.

The team believes their tool could not only be useful in accurately predicting alcohol abuse, but also prove a cost-effective tool for doing so.

"Natural language processing has adequate predictive validity for identifying alcohol misuse in the trauma setting," the authors say. "The technique provides an automated approach to potentially overcome staffing and patient barriers for SBIRT programs at trauma centers."