Use of Epic® to facilitate high-quality randomization of emergency department-based pragmatic clinical trials.

Background: Pragmatic clinical trials are critical to determine the real-world effectiveness of interventions. Emergency departments serve as key clinical sites for such trials because they provide care to large numbers of patients at the earliest point in their hospital encounter. However, few methods that facilitate efficient emergency department-based pragmatic trials have been described. Our objective was to design and embed randomization schemes directly into the electronic health record to facilitate enrollment of emergency department patients by clinical staff 24 h per day, while maintaining concealed and balanced allocation.

Methods: We designed two multi-center pragmatic trials screening for human immunodeficiency virus (HIV TESTED trial) and hepatitis C virus (DETECT Hep C trial). HIV TESTED included four sites, three arms, and compared two forms of risk-assessed (targeted) to non-risk-assessed (nontargeted) screening for human immunodeficiency virus; DETECT Hep C included three sites, two arms, and compared targeted to nontargeted screening for hepatitis C virus. Participant entry occurred during a patient's emergency department visit in conjunction with standard emergency department care. Patient-visit level randomization schemes were designed in each electronic health record system using Epic^® (Epic Systems, Verona, WI; two sites for HIV TESTED; all three sites for DETECT Hep C). Randomization and all forms of screening were fully embedded and integrated in the electronic health record and administered by nurses as part of routine care. Absolute differences and chi-square testing were used to evaluate randomness of study arms and to compare baseline characteristics. Time-motion methods were also used to assess the time of screening and randomization by nurses.

Results: During 61 cumulative enrollment months, 365,462 patient visits occurred. After excluding 184,023 visits by designed electronic health record logic or manual nurse input due to age, previously known human immunodeficiency virus or hepatitis C virus, high acuity or altered mental status, 181,439 patient visits were randomized to one of the interventions. Absolute differences between targeted and nontargeted arms differed by 0.2% and 0.4% (HIV TESTED), and 0.1% (DETECT Hep C), and median absolute differences across all baseline characteristics (i.e. age, sex, race, ethnicity, language, payor, arrival mode, and acuity) between arms were 0.02% (range: -0.7% to +0.5%) and -0.07% (range: -0.3% to +0.7%), and -0.02% (range: -0.7% to +0.7%), respectively. Median time required for nurses to execute randomization ranged from 16 to 67 s, depending on the arm.

Conclusion: Integration of blinded randomization schemes into electronic health record systems resulted in high-volume balanced enrollment of participants 24 h per day while maintaining concealed allocation. Use of this technology is an important tool to scale performance of high-quality pragmatic trials in emergency department settings while minimizing bias.