Better cannabis tracking starts with more transparent health data
Natural Language Processing to Identify Substance Use in Electronic Health Records: A Scoping Review.
AI Summary
This scoping review examined how natural language processing (NLP) is used to identify substance use in electronic health records (EHRs). Across 86 studies, researchers used rule-based systems, conventional machine learning, deep learning, and large language or transformer-based models to detect tobacco, alcohol, opioids, cannabinoids, stimulants, and polysubstance use. Most studies reported performance metrics above 0.80, although the abstract does not provide a single overall accuracy estimate.
Cannabinoid use appeared in only 9 studies, compared with much greater attention to tobacco, opioids, and alcohol. This means NLP tools may be useful for recognizing cannabis-related information in clinical records, but the evidence base is comparatively limited. The review also found limited transparency and reproducibility: only 22 studies published their code, and few shared detailed model specifications or datasets. Better reporting and routine sharing of research materials could improve cannabis surveillance and future clinical research.
π‘ Key Findings
π
Original Abstract
Related Research
Similar Studies
More Addiction research papers you might find interesting.
Review finds strongest evidence in select seizure and MS symptoms
Racism-related distress linked to cannabis problems in Asian American adults
Review maps digital CBT for substance use, but does not establish efficacy
Explore More Research
Stay informed about the latest cannabis science.