Close this panel
Browse By Poster Author
Browse By Title
Close this panel

(551) Predicting Ongoing Opioid Use Among Emergency Department Patients Treated for Back Pain


Kennon Heard, MD – Professor of Emergency Medicine, University of Colorado Denver-Emergency Medicine

Caroline M. Ledbetter, n/a – Sr Professional Research Assistant, University of Colorado, Denver

Jason A. Hoppe, DO – Associate Professor, University of Colorado, Denver

Kennon Heard, MD – Professor of Emergency Medicine, University of Colorado Denver-Emergency Medicine


Background and Objectives: Back pain is a frequently treated with opioids in the emergency department (ED). While opioids offer effective symptom control for acute pain, long-term use is associated with significant negative health effects. The objective of this study is to develop a risk-stratification tool based on variables readily available in the electronic health record (EHR) that be used to identify patients at risk for ongoing opioid use (OOU) following an ED visit for back pain.

Methods: A retrospective study of adult, opioid-naïve (ON) patients discharged from the ED with a diagnosis of back pain. Clinical variables were abstracted from the EHR and opioid prescription fills for the 6 months prior to and following index ED visit were determined from the state prescription drug monitoring program. Patients were considered ON if they had no opioid prescriptions in the 6 months preceding the ED visit and OOU was defined as >90 day supply of opioids filled in the 180 days following index ED visit. We used five algorithmic methods for prediction based on their ability to predict class probabilities well: - logistic regression, classification and regression trees (CART), boosted trees, random and multivariate adaptive regression splines (MARS). A non-informative model that uses no information from predictors was used for comparison (using AUC)

Results: Among 24,487 ON patients, the median (IQR) age was 38 (28-52) years, 55% were women and 56% were non-Hispanic White and 575(2.4%) had OOU. All models performed significantly better than chance. Logistic regression performed the best (AUC 0.70) however the sensitivity and specificity were only 64% and 62% respectively. Age, race/ethinicity, insurance type, pain score, administration of sedatives in the ED and opioid prescription at discharge were most associated with OOU.

Conclusion: While we were able to identify clinical characteristics associated with OOU, we could not develop a prediction tool with a high sensitivity and specificity. Limitations to this study include the use of only prescription opioids in our definition of OOU, the possibility of inaccuracy of the EHR for clinical variables and the lack of generalizability to other settings.

Rate This Poster

Stuff for notes
Stuff for Message board

Share Poster


Technical Support

(877) 426-6323


SUBMIT FEEDBACKfeedback icon

We really appreciate your feedback on the eventScribe website. We use the data to improve the experience and simplify the process for users like you.


Log In / Sign Up

Already have a Schedule or Mobile App?

Don't have an account?

Sign Up

You need to be logged in to bookmark posters, save notes, or rate posters.