Using natural language processing to identify problem usage of prescription opioids.

Area of Science:

• Health Informatics
• Public Health Surveillance
• Natural Language Processing

Background:

• Accurate surveillance of prescription opioid misuse is crucial for public health.
• Traditional diagnostic codes often fail to capture the full extent of problem opioid use.
• Clinical notes contain valuable but often inaccessible information on patient opioid use.

Purpose of the Study:

• To develop and evaluate a method combining NLP and manual review to detect problem opioid use in EHRs.
• To compare the efficacy of this NLP-based method against traditional ICD-9 coding.

Main Methods:

• Utilized EHR data from 22,142 patients on chronic opioid therapy (2006-2012).
• Developed an NLP system with a 1288-term dictionary to identify mentions of opioid misuse, abuse, or overuse in clinical notes.
• Conducted computer-assisted manual review of selected clinical notes for evidence of problem opioid use.

Main Results:

• Traditional ICD-9 codes identified problem opioid use in 10.1% of patients.
• The NLP-assisted method identified an additional 3.1% of patients with problem opioid use.
• High inter-rater reliability (kappa=0.86 and 0.71) was achieved during manual review.

Conclusions:

• Scalable, semi-automated NLP methods efficiently and accurately detect problem opioid use in EHR text.
• Integrating NLP into surveillance can increase prevalence estimates of problem opioid use by up to one-third.
• This approach enhances the understanding and management of prescription opioid misuse.