Background: Emergency departments are increasingly the primary location for healthcare services by individuals at risk or suffering from opioid use disorder (OUD). Electronic medical record (EMR) phenotyping, defined as computerized querying of heterogeneous records, supports the measurement and comparison of interventions capable of increasing the adoption of emergency department-initiated medication for OUD.
Methods: A computable phenotype to identify patients with OUD was developed and evaluated using structured clinical data. Two algorithms were developed: algorithm 1 combined clinician and billing codes, while algorithm 2 used string and character matching to extract data from chief complaints. To evaluate the algorithms, two emergency medicine residents with a third acting as adjudicator reviewed a pragmatic sample of 100 charts meeting at least one algorithm criterion, and assessed for presence of OUD as defined by Diagnostic and Statistical Manual of Mental Disorders, 5th Edition (DSM-5) criteria.<
Results: Agreement between reviewers was high (Cohen’s kappa of 0.95) . algorithm 1 had a Positive Predictive Value (PPV) of 0.96 and Negative Predictive Value (NPV) of 0.98, while for algorithm 2 both values were 1.0. The most frequently met DSM-5 criteria were opioids taken in larger than intended amounts or opioid use in physically hazardous situations, while the least frequent criteria were those describing social dysfunction related to the use of opioids.
Conclusion: This computable phenotype was created for an upcoming pragmatic trial which aims to identify OUD patients eligible for an intervention with high specificity. This phenotype is capable of identifying OUD patients in the EHR, with high predictive value and reliability for the subsequent trial.