A pragmatic study of Clinical Decision Support to promote Prescription Drug Monitoring Program use

JASON A. HOPPE, DO
University of Colorado Anschutz Medical Campus

Providing appropriate, safe analgesia during the opioid crisis a challenge. Prescription Drug Monitoring Programs (PDMPs), an evidence-based intervention to improve prescribing decisions, are underutilized. Electronic health record (EHR) based clinical decision support (CDS) represents a pragmatic, patient-specific and scalable implementation strategy to promote behavior change. Our objective is to assess the development and early deployment of a pragmatic, cluster randomized trial of CDS tools to facilitate PDMP use.

UCHealth system including 14 hospitals with associated Emergency Departments (EDs), 12 free-standing EDs, and approximately 250 ambulatory clinics, approximately 2850 total prescribers, and over 3.6 million total patient visits in 2018.

This is an IRB-approved cluster randomized study. Using published best practices and a literature review, we developed two prototype logic-driven patient-specific CDS to identify patients at risk of opioid abuse/overdose and trigger an interruptive alert to review the PDMP:(1) PDMP risk criteria alone and (2) PDMP + EHR risk criteria. Iterative CDS modifications were informed by interviews with target adopters and organizational decision makers, including clinical observations. Following silent testing of the revised alerts, we randomized providers to receive one of the two alerts or a control alert (fires for all opioid/benzodiazepine e-prescriptions). Alerts were tested live in the academic ED (156 providers and 88k unique patients in 2020) for one month prior to activation in all system EDs. Inpatient and ambulatory settings were activated in a staggered fashion. Education was disseminated through presentations, emails from leadership, a featured article in the system EHR newsletter, and individual provider contacts.

Qualitative feedback from 20 interviews and 10 workflow observations identified concerns for interruptions, alert fatigue, and alert suppression. We completed 3 months of silent testing on ED discharges: control alert triggered 8.1% (95%CI 7.9-8.3%) of discharges, with variability across EDs (5.1-11.3%). PDMP only alert triggered 2.8% (95%CI 2.7-3.0%) before modification and 2.0% (95%CI 1.9-2.1%) after user-driven modification. PDMP + EHR alert triggered 4.2% (95%CI 4.1-4.4%) before modification and 3.7% (95%CI 3.5-3.9%) afterward. Study is ongoing: firing rates, adoption, and PDMP use for all 3 CDS tools in EDs, inpatient facilities and ambulatory clinics will be reported.

User-centered design with key stakeholder input and pilot testing on a large set of target patients helped refine CDS tools for deployment in a pragmatic, cluster randomized trial. Silent alert firing data can inform implementation decisions, validate monitoring tools, and give context to CDS education and messaging. Feedback from the target population can inform study design to improve buy-in.