Computable Structured Phenotype Versus Large Language Model Identification of Opioid Use Disorder Using Electronic Health Record Data

The authors aimed to compare the effectiveness of a rule-based computable phenotype and a large language model in identifying opioid use disorder among patients in the emergency department, using expert physician review as the reference standard. Their findings indicated that while both methods demonstrated strong diagnostic performance, the large language model had significantly higher specificity and positive predictive value, suggesting it may be more effective in minimizing false-positive alerts in clinical settings. Further validation in diverse populations is recommended to confirm these results.