Reader comment regarding machine learning models to detect opioid misuse in emergency department patients at triage

Publications Category : Emergency Medicine

Author(s): Christian Angelo I. Ventura, Ava D. Cabanilla, Edward E. Denton

Project Description : Chhablani and colleagues make an important contribution by demonstrating that machine learning classifiers applied to triage-level electronic health record data can identify emergency department (ED) patients with opioid misuse prior to physician evaluation [1]. Gradient boosting and random forest models achieved F1 scores of 0.77 and 0.75 and area under the precision-recall curve values of 0.89 and 0.88 respectively, performance comparable to opioid-related ICD-10-CM codes placed later in the encounter. Given that EDs represent a critical but underutilized contact point for patients with opioid use disorder, and that evidence-based interventions such as buprenorphine initiation and naloxone distribution remain dramatically underutilized, the case for automated triage-based screening is compelling. I offer three considerations as this work moves toward prospective validation and deployment.