Automated Technique Identifies Problematic Opioid Use From EHR Notes
A machine learning approach that automated the Addiction Behaviors Checklist identified problematic opioid use in patients with chronic pain by analyzing electronic health record (EHR) notes, according to a study published in JAMA Psychiatry.
“Electronic health records allow large-scale studies to identify a continuum of problematic opioid use, including opioid use disorder. Traditionally, this is done through diagnostic codes, which are often unreliable and underused,” explained corresponding author Alvin D. Jeffery, PhD, of the Vanderbilt University Medical Center Department of Biomedical Informatics, Nashville, Tennessee, and coauthors in the study background.
READ>>Low-Dose Buprenorphine Initiation May Prevent Opioid Withdrawal in Some Patients
The cross-sectional study tested the performance of an interpretable natural language processing technique that automated the validated Addiction Behaviors Checklist to identify problematic opioid use in deidentified EHRs for 8063 patients with chronic pain from Vanderbilt University Medical Center. Researchers tested their model against a blinded, manually reviewed holdout test set. Afterward, the automated approach was externally validated against an independent test set of 100 patients with chronic pain from Geisinger.
“In this cohort of patients with chronic pain, the automated approach achieved high sensitivity and positive predictive value compared with a manual review and performed significantly better than diagnostic codes,” researchers reported.
According to the study, the sensitivity and positive predictive value, measured by F1 scores, was 0.73 with the automated approach and 0.08 for diagnostic codes in the Vanderbilt cohort. In the Geisinger cohort, the F1 score was 0.70 for the automated approach compared with 0.29 for diagnostic codes.
Meanwhile, the area under the curve (AUC) in the Vanderbilt cohort was 0.82 with the automated approach compared with 0.52 with diagnostic codes. The AUC was 0.86 with the automated approach compared with 0.59 with diagnostic codes in the Geisinger cohort.
“The automatic characterization of problematic opioid use from existing clinical notes could be transformative for preventive care of chronic pain and surgical patients and identification of patients with probable opioid addiction, which would address a care gap in opioid screening practices,” researchers wrote.
