Real-Time Identification of Opioid Use Disorder in the Emergency Department Using a Machine-Learning Phenotype

EDs continue to face challenges identifying Opioid Use Disorder (OUD) in real time, contributing to the underuse of buprenorphine (BUP). Addressing this gap requires real-time identification and phenotyping of OUD patients to improve treatment initiation. The primary objective is to develop and deploy a real‑time, EHR‑integrated machine learning phenotype to identify ED patients with OUD for buprenorphine initiation.

We conducted a multi‑phase study across three EDs in a single United States health system from 2014 to 2025. Using visit‑level data available at or before triage, we trained a random‑forest classifier with an 80:20 train–test split and cross-validation to estimate OUD risk and embedded scoring in the EHR to trigger point‑of‑care alerts. A computable silver‑standard label based on the diagnosis code supported retrospective model development; a clinician-adjudicated, gold-standard reference for model evaluation was established via structured, DSM-5–aligned chart review conducted independently by two physicians. Performance was summarized with ROC‑AUC, calibration, and threshold‑based classification metrics; prospective validation used a stratified random sample of flagged and unflagged encounters.

Retrospective discrimination compared to the silver standard was high (n=1,336,335, ROC‑AUC 0.99, 95% CI 0.98-0.99), and calibration plots informed operating‑point selection for real‑time use. In prospective gold‑standard validation (n=218, Cohen’s kappa = 0.82), the positive predictive value was 98.28%, and the negative predictive value was 95.68% at the prespecified threshold. The model yielded a sensitivity of 0.96 (95% CI 0.90 - 0.98) and a specificity of 0.91 (95% CI 0.85 - 0.95) with respect to the gold standard.

An EHR‑embedded, machine learning phenotype can accurately and feasibly identify ED patients with OUD in real time. Ongoing work will report operational metrics, monitor performance drift and equity across subgroups, and evaluate downstream clinical outcomes.