A Naloxone Best Practice Advisory That Failed to Signal: A Real‑World Evaluation of an Emergency Department

BACKGROUND: Naloxone distribution is a key harm-reduction strategy, yet implementation in emergency departments (EDs) remains inconsistent. Best Practice Advisories (BPAs) are intended to provide timely, actionable prompts, but high alert burden can diminish their effectiveness through alert fatigue. Existing evidence largely reflects academic medical centers and focuses on naloxone prescribing rather than the distribution of take-home naloxone (THN) kits. Little is known about how naloxone-related BPAs perform in community EDs, where workflows and alert-fatigue dynamics differ. METHODOLOGY: We conducted a retrospective cross-sectional study in a suburban, non-academic ED from January 2022 to January 2024. After exclusions, 10,313 encounters with complete BPA and naloxone documentation were analyzed. Six naloxone-related BPA elements were extracted from audit logs and evaluated alongside demographic and clinical variables. Multicollinearity among BPA components was addressed using variance inflation factor (VIF)-guided pruning. Class imbalance was mitigated using SMOTE-Tomek and class weights. Logistic regression models were developed using an 80/20 train-test split, with performance assessed using area under the curve (AUC), accuracy, and sensitivity analyses. RESULTS: Among encounters in which a naloxone-related BPA fired, clinicians overrode 25.6% of alerts and acknowledged 74.4%; however, acknowledgment translated into THN distribution in only 36.7% of encounters. This highlights a substantial gap between alert visibility and clinical action. Furthermore, BPA-related variables contributed minimal predictive signal. In adjusted models, the Emergency Severity Index (ESI) was the only significant predictor, with lower acuity associated with decreased odds of naloxone distribution. Race, ethnicity, sex, age, and all BPA-related variables were not significant predictors. The final model demonstrated moderate discrimination (AUC ≈ 0.77) with stable performance across original and SMOTE-balanced datasets. Subgroup analyses showed no evidence that BPAs improved equity in naloxone distribution. CONCLUSIONS: Naloxone BPAs in this community ED did not function as meaningful clinical signals. Although most alerts were acknowledged, they did not influence the distribution of THN kits and added minimal predictive value. Naloxone distribution reflected clinical cues, particularly acuity and opioid-related risk indicators, rather than BPA prompts. Redesign is needed to improve the timing, targeting, and relevance of naloxone-related clinical decision support (CDS). Future harm-reduction strategies should prioritize workflow-integrated, actionable CDS rather than broad, interruptive alerts that blend into background noise.