Predictors of emergency department opioid administration and prescribing: A machine learning approach

Introduction: The opioid epidemic has altered normative clinical perceptions on addressing both acute and chronic pain, particularly within the Emergency Department (ED) setting, where providers are now confronted with balancing pain management and potential abuse. This study aims to examine patient sociodemographic and ED clinical characteristics to comprehensively determine predictors of opioid administration during an ED visit (ED-RX) and prescribing upon discharge (DC-RX). Methods: ED visit data of patients ≥18 years old from the National Hospital Ambulatory Medical Care Survey (NHAMCS) from 2014 to 2017 were used. Opioid prescriptions were determined utilizing Lexicon narcotic drug classifications. Visit characteristics studied included sociodemographic variables, and ED clinical variables, such as chief complaint, and discharge diagnosis. Machine learning methods were used to determine predictors of ED-RX and DC-RX and weighted logistic regressions were performed using selected predictors. Results: Of the 44,227 ED visits identified, patients tended to be female (57.4%), and White (74.2%) with an average age of 46.4 years (SE = 0.3). Weighted proportions of ED-RX and DC-RX were 23.2% and 18.9%, respectively. The strongest predictors of ED-RX were CT scan ordered (OR = 2.18, 95% CI = 1.84–2.58), abdominal pain (OR = 1.93, 95% CI:1.59–2.34) and back pain (OR = 1.81, 95% CI:1.45–2.27). Tooth pain (OR = 6.94, 95% CI = 4.40–10.94) and fracture injury diagnoses (OR = 3.76, 95% CI = 2.72–5.19) were the strongest predictors of DC-RX. Conclusions: These findings demonstrate the utility of machine learning for understanding clinical predictors of opioid administration and prescribing in the ED, and its potential in informing standardized prescribing recommendations and guidelines.