Background: Opioid overdose (OD) a leading cause of accidental death in the US. Multiple community initiatives have been implemented for the prevention opioid overdose (OD), and novel approaches to identify patients who present to the emergency department (ED) with high risk for opioid overdose using electronic health records (EHR) would be extremely important. However, these patients can be difficult to capture using standard methods . The purpose of this study was to develop a method of identifying a cohort of patients at risk for opioid OD using EHR to identify those what are at high risk for a subsequent OD.
Methods: This study was performed in a major urban Midwestern hospital system. After IRB approval, all ED visits during October 2018 were reviewed. An EHR algorithm and was created using eight search elements: 1) reason for visit, 2) reason for visit specifically OD, 3) reason for visit comments including opioids or heroin or Narcan, 4) naloxone given in the ED, 5) suboxone ordered from the ED, 6) diagnosis, 7) ED peer recovery coach “smartphrase”, and 8) peer recovery coach search in the provider notes. Next, patients charts were explicitly reviewed by the investigators to identify the cohort of 91 patients who would be discharged from the ED with a high risk of OD. Those included patients presenting with an OD, patients in acute withdrawal, patients with suicidal ideation secondary to their opioid abuse. A scoring system was then created to using these elements to identify a highly specific cohort of patients at risk for subsequent OD. Criteria was divided into highly suggestive (diagnosis, suboxone ordered), and suggestive (remainder of criteria). Patients discharged with one highly suggestive or more than two suggestive criteria were considered high risk for OD by the scoring system.
Results: This search identified 292 unique patients during the one-month time span. The scoring system correctly identified 79/91 patients deemed high risk for opioid OD and correctly identified 199/206 patients not considered high risk. The sensitivity was 86% (95% CI 78-93%) and the specificity was 97% (95%CI 94-99%).
Conclusion: Novel search methods using electronic health records can identify a cohort of patients discharged from the ED with a high risk of opioid OD. By identification of this cohort, this approach will allow rigorous analysis of programs to treat opioid OD.