This report follows the SQUIRE 2.0 standards for QI reporting.20 Details on the routine rollout, QI protocol development, components and implementation are described elsewhere.19 The Institutional Review Board deemed the project a QI initiative, not constituting human subjects research, as defined under 45 CFR 46.102(d).
Context and Setting The academic health system included 35 primary care (internal medicine and family medicine) clinics in rural, suburban and urban settings, caring for 200,000 adult primary care patients. In February 2016, it introduced a guideline-driven opioid policy on the management of long-term (≥ 3 months) opioid therapy in primary care adult outpatients with chronic non-cancer pain (“target patients”). The opioid policy provided multiple guideline-concordant recommendations for safety, treatment response monitoring, and management in this target population.19 The system-wide routine rollout of this policy involved a pilot-testing of the rollout methods in 3 clinics in the fall of 2015, followed by the system-wide rollout in February 2016 that consisted of an one-hour in-person meeting for clinicians; an one-hour online training session for clinic staff; and two follow-up tele-meetings to address comments/questions from the clinic staff.19The health system’s leadership approved this QI project to determine if augmenting the routine rollout would improve outcomes in this target patient population.
Clinic Selection The health system included 35 primary care clinics. Nine of these clinics were involved in other opioid QI initiatives and were excluded from the recruitment pool, yielding 26 clinics eligible for the QI project and included in the analyses. Among these 26 clinics, those with the highest number of target patients were approached first. Three of the approached clinics declined (“too busy for new QI efforts”). The first 9 consenting clinics (convenience sample) were enrolled into a non-randomized stepped-wedge QI project. The QI intervention was initiated immediately after the health system’s routine policy rollout in February 2016. It was delivered over 4-6 months at each clinic and implemented in three waves (March-July 2016; September-December 2016; January-June 2017), with 3 clinics per wave; the QI wave assignment was non-random, based on each clinic’s preference.
Participants The intervention subjects were volunteer clinical staff (prescribers, nurses and others) at each intervention clinic. The evaluation subjects (target patient population) were identified by the search of EHR-based data from the problem list, encounter, and billing records,using the health system-developedcriteria: age ≥ 18 years old; active-patient status (seen at the clinic in the past 3 years); primary care provider within the health system; no diagnosis of malignant neoplasm (except non-melanoma skin cancer) or palliative or hospice care status; and meeting at least one of the two criteria: 1) ≥1 opioid prescription issued in the prior 45 days and ≥3 opioid prescriptions issued in the prior 4 months; or 2) ≥1 opioid prescription issued in the prior 45 days, and presence of a chronic pain diagnosis and a controlled substance agreement. For the analyses, buprenorphine was excluded from the “eligible opioid” list due to its primary utility locally as a treatment for opioid use disorder.
QI Intervention The intervention was developed by the project’s multidisciplinary team with input from the health system leadership and clinicians to ensure alignment with opioid prescribing guidelines and the health system’s policies, procedures and resources, as detailed elsewhere.19 Briefly, the QI intervention at each clinic consisted of: a) one 1-hour academic detailing session, delivered by the project physicians, outlining the project, the national guidelines, and the health system’s opioid policy recommendations; b) two 20-21 question online educational modules: one on shared decision-making in the context of opioid therapy for chronic pain, and another on the guideline and health system policy recommendations for opioid therapy management; and c) six 1-hour practice facilitation (PF) sessions delivered at each clinic over 4-6 months by the project’s trained facilitators. The PF sessions focused on optimizing clinical workflows to promote clinician adherence to the guideline and health system policy recommendations with measurable outcomes (“QI targets”). The selection of QI targets for the PF sessions was driven by each clinic team’s preference.19 Participating clinical staff were eligible for up to 23 AMA PRA Category 1™ Credits for completing the intervention.
Measures De-identified outcome measures were extracted monthly from the EHR (Epic Systems Corporation) from baseline (January 2016) through project end (December 2017). Data were analyzed at the clinic level. Outcome measures were selected based on the opioid prescribing guideline15-17 and policy recommendations, and availability of EHR data entered as a part of health system’s routine care.19 The percentage of target patients with a “current” (signed within the past 12 months) treatment agreement was chosen as the primary outcome based on the health system’s opioid policy, which recommended its routine completion, followed by annual updates.19 Secondary outcome measures were: a) “current” (past 12 months) urine drug testing (UDT) and b) depression screen with a two- or nine-item Patient Health Questionnaire (PHQ)21,22 (a positive screen using a two-item questionnaire automatically triggered completion of a nine-item questionnaire); c) prescription drug monitoring program (PDMP) database check; d) completion of the opioid misuse risk screen with the Diagnosis, Intractability, Risk, Efficacy (D.I.R.E.) tool;23 and e) the rate of opioid-benzodiazepine co-prescribing in at least one of the prior 3 months. Secondary outcomes a-d were based on the UWH policy. The co-prescribing measure was not a part of the health system’s policy, but was included based on guideline recommendations advising against co-prescribing.15-17 Additional measures of interest commonly used to assess opioid interventions15-18 included: a) percentage of target patients relative to the total adult clinic panel; b) daily opioid dose, calculated per target patient as an average morphine-equivalent dose (MED; milligrams/day) by adding up the doses of all opioids (except buprenorphine) prescribed for outpatient treatment in the prior 90 days and dividing the sum by 90; and c) percentage of target patients prescribed MED ≥90 mg/day (past 90 days). Opioid and benzodiazepine prescription data were extracted from the medication list.
Statistical analysis Nine clinics were predicted to provide 80% power to detect a clinically meaningful (20%) increase in the use of a treatment agreement (see Additional File 1 for a detailed sample size discussion). Primary and secondary outcomes were defined a priori. SAS Version 9.4 was used for statistical analyses. Baseline (January 2016) and project end (December 2017) data were collected using clinic-level averages, weighted by target patient panel size per clinic. Descriptive statistics were used to describe these data. Single and two-sample means tests evaluated outcome changes between baseline and exit data within and between the intervention and comparison clinics, respectively. For the PDMP outcome measure, the end date was changed from December to March 2017 due to changes in state law and health system requirements, which led to approximately 100% PDMP check documentation across the clinics starting in April 2017. The primary evaluation of intervention impact was conducted using a mixed-effects regression analysis model. The model leveraged the monthly EHR data and accounted for the timing of intervention delivery in the intervention clinics by contrasting clinic-level pre-intervention data, with data collected during and then after the intervention (stepped-wedge analysis). The stepped-wedge analysis was further augmented by adding comparison clinics’ monthly data during the same assessment period. Linear curves were fitted to the monthly outcomes as fixed effects, with baseline values and slopes of change separately estimated for intervention and comparison clinics. Additional fixed effects were included to allow the slopes of fitted curves for intervention clinic outcomes to change in relation to the intervention and post-intervention periods. Random effects were included at the levels of both the primary care provider (PCP) within each clinic and the clinic as a whole to account for correlation among monthly observations from the same PCP or the same clinic. Observations were also weighted by the number of target patients within each PCP’s monthly panel. Estimates of the differential slopes (pre-intervention, intervention and post-intervention) for the intervention clinics and a single, study-long slope for the comparison clinics were used to assess the impact of the intervention on the intervention clinics’ outcomes. Baseline differences between intervention and comparison clinic characteristics, and between clinic and PCPs within each study group, were accounted for in differential baseline intercepts and slopes, and with random intercept and slope effects, respectively. See Additional File 2 for details of the mixed effect model and result interpretation.
A subgroup analysis was conducted among target patients treated with MED ≥90 mg/day, because of their increased risk for opioid-related harm.18
The significance and magnitude of changes were assessed with p values (significance level: two-tailed p<0.05), 95% Confidence Intervals (CIs), and/or Cohen’s d (ES, 0.2-0.4: small; 0.5-0.7: medium; ≥0.8: large effect size).