医疗卫生系统干预以减少阿片类药物处方过量前后开此处方行为的评估

Importance  Overprescribing of opioids has generated and sustains the opioid overdose epidemic. Health systems have a responsibility to lead the effort to reduce overprescribing.

Objective  To measure the effects of multilevel interventions on opioid prescribing within a health system.

Design, Setting, and Participants  Quality improvement study comparing a 6-month preintervention baseline with a 16-month postintervention period ending in April 2018. Inpatient and outpatient clinical activity within a regional health system including an acute care hospital, same-day surgery, and outpatient clinics. Opioid prescribing activity by hundreds of clinicians involving over a million clinical encounters was measured using a health system’s electronic medical record.

Interventions  Multiple parallel interventions in different domains, including prescriber education and accountability, enhanced oversight via measurement of individual prescribers, tools to right-size postoperative discharge prescriptions, reduction of default amounts on standard opioid prescription orders, and professionally written patient and public education about opioid risks and alternatives.

Main Outcomes and Measures  Morphine milligram equivalents (MME) per encounter per month, MME per opioid prescription, and rate of opioid prescriptions (opioid prescriptions per encounter per month).

Results  More than 44 000 clinical encounters per month were recorded. All baseline trends were not significantly different from 0. Total health system MME per encounter decreased 1.0 MME per encounter per month. At the end of the postintervention observation period, the monthly MME per encounter was 58% lower than the average of the 6-month baseline, the MME per opioid prescription per month was 34% less than the average of the baseline, and the opioid prescription rate was 38% lower than the average of the baseline.

Conclusions and Relevance  Opioid overprescribing was reduced with multifocal interventions targeting patient and public demand, creating prescriber awareness and accountability, and creating tools for clinical leadership accountability. The interventions described are adoptable by most organized health systems. Reducing total opioid supply within communities should be given high priority by those with a mission to protect and improve public health.

The rapidly increasing death rate in the United States from opioid overdoses has attracted national and international attention in both clinical circles and in the public space.¹ States of emergency have been declared both at the federal level and in local jurisdictions. A diverse array of solutions have been proposed, including criminal justice procedures, health policies, funding allocations, and reforms in clinical practice.^2-4 The rapid growth in opioid overdose deaths, mainly from so-called street drugs such as heroin, fentanyl, and fentanyl derivatives, was preceded by a 300% expansion of retail opioid prescribing beginning in the early 1990s and peaking in 2012.⁵ But even with a slight decline since 2012, prescribing far exceeds 1991 levels and that of other nations.⁶ The causes of this massive prescribing growth have been much debated.^7-9 But it is not debatable that prescribed opioids, including diverted prescription opioids, were the initial source of opioids for most current heroin users.¹⁰ The prescriber’s role in generating and sustaining opioid abuse has been made clear by studies that link a practitioner’s prescribing patterns to a patient’s likelihood of long-term opioid dependence.^11,12 This linkage between prescribing patterns and opioid dependency^7-12 formed the rationale for this targeted initiative to reduce opioid prescribing locally.

In response to the local public health emergency and declared state of emergency, the authors initiated a pilot review of prescribing among physicians, physician assistants, and nurse practitioners in primary care and emergency departments (EDs) using the health system’s electronic medical record (EMR) to collect data on prescribing frequency and overall opioid volume. The pilot analyses revealed high variability in prescribing among practitioners and other deficiencies in management of chronic opioid users, providing the rationale for a comprehensive campaign. The specific aims were to reduce unwarranted variability in opioid prescribing; standardize prescribing for clinically similar situations; reduce total opioid prescribing; create accountability from prescribers and medical directors by collecting and sharing prescribing data from the EMR; and observe for adverse effects by monitoring patient satisfaction with pain management and return visits to the ED. This report describes the initiatives undertaken and the results of a physician-led campaign to reduce opioid prescribing within a health system.

This report conforms to the Standards for Quality Improvement Reporting Excellence (SQUIRE) reporting guideline for health care safety innovations. The Clinical Research Committee of Anne Arundel Health System reviewed this project and determined that it qualified as quality improvement and not research. It also performed an ethics review and did not identify any concerns in initiating the interventions.

Anne Arundel Medical Center is a 385-bed acute care hospital in Anne Arundel County, Maryland, serving as a regional referral center for more than 1 000 000 individuals. Relevant volumes for 2017 include 25 567 medical admissions, excluding newborns, 96 118 visits to the ED, 17 791 inpatient surgical procedures, 6871 ambulatory surgical procedures, and 331 554 000 visits to ambulatory health system clinicians. The health system directly employs more than 326 physicians, nurse practitioners, and physician assistants, including 42 primary care practitioners in 15 separate locations. All clinical activity uses the health system EMR Epic 2015.

A standing report was created that identified all prescription opioids and converted the prescription dose into morphine milligram equivalents (MME) using a standard reference table.¹³ The prescription was then assigned to individual prescribers and organized by particular clinical department or service: primary care, ED, acute care hospitalist, medical specialists, orthopedic clinic, oncology, palliative care, and surgery. Opioid prescriptions from oncology and palliative care were excluded from the analysis as public policies did not seek to change opioid prescribing for patients with advanced cancer or patients in palliative care for painful conditions. The health system employs no pain management specialty physicians. A few iterations were required until prescribers were accurately assigned to specific clinical groups. Individual service reports, which updated daily, were built for medical directors allowing comparison among prescribers. The reports included patient medical record numbers so that responsible medical directors could review the medical record to determine the clinical circumstances of the prescription.

Prescribing trends were overseen by a newly created Opioid Prescribing Task Force composed of clinical and administrative leaders from pharmacy, mental health, surgery, orthopedics, primary care, ED, community health, informatics, quality, and health system administration.

A multilevel campaign was initiated comprising large group education that included departmental grand rounds, service meetings with data review, and circulation of medical journal articles with information on overprescribing. A visual display of individual clinician prescribing for ready comparison with service peers was created. This was followed by academic detailing¹⁴ by medical directors with one-on-one meetings with prescribers to reinforce the key points of the education and review individual prescribing data and comparison with peers. The topics and main points of the education efforts are shown in eTable 1 in the Supplement. The principal goal was to convey an appreciation for the idea that medicinal opioids generated and sustain the opioid overuse epidemic and that prescriber behavior influences patient outcomes with regard to opioid dependence.

Although all members of the medical staff were included in the education efforts and had access to all tools introduced, 3 clinical areas, the ED, primary care, and the orthopedic clinic, were chosen for initial focused efforts because of high opioid prescription rate and volume. Some clinical departments, such as the ED and upper extremity orthopedic surgeons, rapidly adopted service-specific standardized approaches to prescribing following guidance to use the smallest amounts of opioid that might still be effective.⁴ Other departments, such as obstetrics and surgery, created these standardized approaches during the course of the intervention year.

New EMR-based tools included patient education on opioid alternatives and opioid safety, identification of patient-level characteristics associated with elevated risk for dependence,¹⁵ facilitated referral for substance abuse counseling, and detoxification within the EMR. An easy-to-use tracking tool for opioid consumption during an inpatient stay was created to help reduce mismatched patient need and discharge prescription amounts, a problem that has been cited as a cause of overprescribing.¹⁶

The goal was to enhance the public’s level of caution about opioids and thereby reduce demand for unnecessary opioids in any potential medical encounter. Professionally developed written materials from the Anne Arundel County Department of Health in English and Spanish were reviewed with a cadre of patient advisors. The materials were used to communicate with the public about risks of opioid use, alternatives to opioids, safe use, safe storage, and disposal. Messages were delivered through various communication channels available to the health system, including local and regional radio and television interviews with physician leaders, authored columns in local newspapers, website articles, social media postings, and signage within the medical center.

As above, the goal was to enhance the patient’s level of caution about opioids and achieve reasonable pain management goals, particularly after surgery. Similar materials and processes, including use of patient advisors to review materials, were used to develop materials that support the use of nonopioid alternatives for acute or chronic pain. New patient safety instructions were created to accompany every EMR-generated opioid prescription in the after-visit summary, including safe storage and disposal. The instructions included explanation of side effects, including emphasis on risk of sedation and respiratory depression, warnings about concurrent sedating medications and alcohol, safe storage, and disposal of unused medication. Discharge instructions also included the addresses of medication take-back programs. Revised patient education materials regarding appropriate pain management goals were created. eTable 2 in the Supplement shows the time course for initiation of the interventions, all of which were maintained and expanded as interventions disseminated to additional departments, surgical procedures, and services.

The study measured all ambulatory and hospital discharge opioid prescriptions, including same day-surgery, before and after intervention. Patient satisfaction with pain management was measured through an existing survey, the Clinician and Group Consumer Assessment of Healthcare Providers and Systems (CG-CAHPS), to observe for adverse trends in patient satisfaction after the interventions were initiated. Because national data suggest a modest existing trend toward reduction in overall prescribing,⁶ we included a 6-month preintervention data view beginning in April 2016 to assess any secular preintervention trend already under way. Interventions were phased in over a 3-month period beginning in October 1, 2016. The postintervention observation period began on January 1, 2017, and continued through April 30, 2018. To account for any changes in clinical volumes that might affect opioid volume, we collected data on number of encounters using standard health system reports.

The volume and rate of opioid prescribing through the EMR were the key outputs measured. Data were analyzed as MME per patient encounter per month, MME per prescription, and number of opioid prescriptions per encounter.

Separate linear regression analyses were performed for the parameters of aggregated MME per encounter per month, MME per opioid prescription, and rate of opioid prescription per encounter for the 6-month baseline period and for the 16 months of postintervention observation. To avoid contaminating either analysis, the 3 months of intervention phase-in were not included in either data set. All analyses were conducted with Bonferroni confidence interval control of type I error rate (0.05 ÷ 6 = 0.00833). The slopes of the 2 linear regressions, before and after intervention, were calculated. P values on the slopes of the preintervention and postintervention trend lines were calculated to determine statistical significance. Two-sided P values of less than .05 were considered statistically significant.

More than 44 000 clinical encounters per month were recorded. The interventions were associated with a reduction in prescribed MME per encounter, MME per prescription, and opioid prescriptions per encounter. All differences in slopes were significant using the Bonferroni confidence interval control method.

Figure 1, Figure 2, Figure 3, and Figure 4 show prescribing trends for the baseline 6 months and for the 16 months after the interventions. In all cases, the baseline trend was not statistically different from 0 and the postintervention negative slope was highly statistically significant at P < .001. Total health system MME per encounter decreased 1.0 MME per encounter per month from a starting level of 34.4 MME per encounter per month (Figure 1). At the end of the postintervention observation period, the monthly MME per encounter was 58% lower than the average of the 6-month baseline. Morphine milligram equivalents per opioid prescription decreased 8 MME per prescription per month from a starting level of 428 MME per prescription per month (Figure 2). At the end of the 16-month postintervention period, the MME per prescription per month was 34% less than the average of the baseline.

The percentage of health system clinical encounters resulting in an opioid prescription was reduced by 0.2% each month from a starting level of 11.5% (Figure 3). At the end of the 16-month postintervention period, the opioid prescription rate was 38% less than the average of the baseline.

Analysis of MME per encounter per month for the 3 initial targeted clinical departments (Figure 4) shows improvement in all 3. Primary care MME per encounter decreased 0.8 from a starting point of 34.8, resulting in a 55% reduction from baseline. Emergency department prescribing decreased 0.4 MME per encounter per month from a starting level of 17. Overall reduction from the average of the baseline 6 months was 49%. Orthopedic clinic MME per encounter decreased 0.9 MME per month from a starting point of 43 MME per encounter. Overall reduction from baseline was 61%.

Variability in MME per encounter was reduced in all 3 targeted areas as measured by the interquartile range (IQR) (Figure 5). In the ED, the IQR decreased from 72 to 29.5 MME per encounter with the number of high outliers decreasing from 5 to 2. Among primary care prescribers, the IQR decreased from 686 to 601 MME per encounter with the number of outliers decreasing from 4 to 0. In the orthopedic clinic, the IQR decreased from 302 to 143 MME per encounter with 1 high outlier in each time period.

Patient satisfaction with pain management in the ED, as measured by the percentage of CG-CAHPS survey respondents who graded their satisfaction as either a 9 or 10 of 10 possible points, did not decrease but rather improved slightly after the interventions from 52% to 61% (eFigure in the Supplement). There was no increase in number of patients who had return visits to the ED within 30 days for pain control after either a surgical procedure or a previous ED visit.

Health systems and their physician leaders have a special responsibility to organize the clinical response to the opioid crisis. Their clinical sites are the source for many opioid prescriptions, they employ or contract with many clinicians both in hospitals and in the community, they are most often the conveners of organized medical staff quality improvement, they have a mission to improve and protect the public’s health, and they have existing channels of communication to educate and inform the public and patients about opioid risks and alternatives.

The success in reducing and standardizing opioid prescribing described in this report is notable for the size and rapidity with which major changes occurred and the large number of clinicians who created the change. Changing physician ordering behaviors on a broad scale is challenging. Successful efforts have been reported with antibiotic use,¹⁷ laboratory test ordering,¹⁸ and blood transfusions,¹⁹ although sustainability of performance improvement is often not demonstrated and there is the potential for submission and publication bias for positive results. Other initiatives, such as efforts to effect cultural change with regard to the ordering of low-value tests, have not succeeded on a broad scale, at least in the early years.²⁰

Our data demonstrate that it is possible, through a coordinated, multilevel campaign, to reduce opioid overprescribing without worsening overall patient satisfaction. The intervention was associated with improvements in multiple measures of opioid prescribing: MME per encounter, MME per prescription, percentage of encounters with opioid prescriptions, and variability within 3 targeted departments. The improvements expanded over the observation period as additional departments and procedures initiated prescribing reforms such as postoperative standardization. The success of the current initiative is likely due to the timely union of public and clinician concerns fostered by strong media and government focus on the opioid overdose crisis. In our experience, cautious prescribing has been welcomed by the broad public and engendered little resistance from most patients.

Initial physician concern about the impact of lowered opioid prescribing and patient satisfaction was addressed with both internal and published data showing patient satisfaction with pain management after surgery is not correlated with the volume of prescribed opioids after postsurgical hospital discharge.²¹ Early adopters of conservative prescribing relayed anecdotes of patients expressing gratitude for the discussion about nonopioid alternatives, thus encouraging other clinicians. Another early physician concern regarded the potential loss of prescribing autonomy. But there were no mandated restrictions added into the EMR during the time of this project, although we did add guidance, transparency, internal benchmarking, and accountability. After initial concern about data accuracy, prescribers came to understand that the gravity of the overdose crisis warrants more oversight by clinically informed medical directors. The visual management comparisons were well received, and many expressed surprise to see where they ranked. In some cases, such as the ED and hospitalists, clinicians expressed satisfaction with department-wide standards that made it easier for them to cite adherence to protocols to reduce prescribing volumes.

The study had a uniform, reproducible, and validated way of measuring opioid prescribing volume. There was a stable population of prescribers numbering several hundred and the work involved hundreds of thousands of patient encounters. The minimal secular trend in reduced opioid prescribing was measured and accounted for.

A few published quality improvement projects have shown a reduction in opioid prescribing with efforts to standardize and monitor prescribing patterns involving either a single clinical area^22-25 or broadly throughout a health system.^26,27 In general, these efforts involve adoption of guidelines and monitoring of adherence with data collection. A retrospective 5-year analysis of the impact of standard statewide prescribing guidelines for ED clinicians in Ohio demonstrated a 39.7% decrease in total number of prescriptions and a 28.3% decrease in prescriptions of more than 3 days of opioids.²⁴ Within the Veterans Health Administration, an opioid prescribing dashboard was created that allowed facility medical leaders to audit facility-level, prescriber-level, and patient-level prescribing. In the year of follow-up after the advent of the monitoring tool in October 2013, there was an acceleration in the already downward trend in overall opioid prescribing (0.4% per month), high-dose (>100 MME) opioid prescriptions (0.67% per month), and patients receiving concurrent opioids and benzodiazepines (0.86% per month).²⁶ A study from Kaiser Permanente Southern California reported the results of a series of interventions rolled out over 6 years that included opioid prescribing data collection and dissemination, dose and duration restrictions, computer decision support meant to discourage long-acting opioids, and warnings against concurrent prescription of benzodiazepines and other high-risk medications.²⁷ Improvements were seen in all targeted metrics.

This study was conducted in a single health system with a single EMR that captured all electronic prescriptions. Strong physician leadership helped drive the program. Other health system resources, including marketing and communications, information technology, and medical informatics helped facilitate rapid changes. Other health systems with less overprescribing might see less impressive results on a percentage basis. However, our county is only a moderate prescribing county according to Centers for Disease Control and Prevention data,²⁸ so similar gains may be possible in many other jurisdictions. Moreover, the specific tools developed are available to almost any health system or practice group with an EMR.

These results are unlikely to be due to concurrent legislative requirements or other external policies, as the mandatory review of a statewide prescription database had not yet taken effect at the time of these initiatives. Enforced limitations of quantity and preauthorization for high doses among Maryland Medicaid patients did occur halfway through our intervention period, but less than 3% of our encounters are with Medicaid recipients, and these restrictions were modest.²⁹

The study lacks prospective controls, as it was not possible to segregate certain clinical areas to avoid the interventions. However, because 6 months of data prior to the interventions showed no statistically significant downward trend, it is likely that the interventions were responsible for the results. Data from the Centers for Disease Control and Prevention showing a much more modest 5.8% reduction in our county, from 74.1 to 69.8 opioid prescriptions per 10 000 population from 2015 to 2016, further support this conclusion.²⁸ National data comparing 2017 with 2016 show a reduction of only 12%.³⁰

The study is further limited by the fact that, similar to other published reports on opioid prescribing reduction,^22,23,27 we did not have research access to state or regional prescription drug monitoring programs. Thus, we cannot absolutely exclude the possibility that our significant reduction in internal opioid prescribing resulted from migration of patients to other practitioners not using our EMR. Mass migration out of the health system seems unlikely given the rapidity and scale of the changes noted here and the fact that most of our clinical volumes increased during the intervention year. Finally, our method of studying patient satisfaction with pain management using the CG-CAHPS tool may have failed to detect patient concerns or dissatisfaction among small groups of patients.

Because the solutions were initiated as a bundle, it is impossible to determine which interventions were most effective. The magnitude of the achieved reduction stands out in comparison with other less successful efforts in which either policies alone³¹ or data sharing without follow-up were used.^32,33 Our program involved 4 key elements, all of which may be necessary for optimum results: patient and family education to reduce demand, achieving prescriber understanding that medicinal opioids are part of the opioid crisis, individual and aggregate data sharing, and oversight by accountable medical directors.

Some observers have noted the trend toward restrictive prescribing and cautioned about too severe a course correction.^34,35 We share these concerns, but our data did not support the presence of large numbers of dissatisfied patients. Indeed, care was taken to create referral pathways for detoxification programs for those patients interested in addressing long-term habits.

The process of standardizing prescribing after surgery still needs to be developed fully. While overprescribing is widely acknowledged,³⁶ the ideal discharge opioid volume for various procedures remains uncertain. Some have argued that the opioid consumption during the last 24 hours before discharge is the best way to determine prescription volume following discharge.³⁷ Indeed, enhanced access to this information was 1 of the tools used in this project. The ability to electronically prescribe opioids remotely through the EMR will facilitate smaller-volume initial prescriptions because a second one can be provided more easily and only if needed.

Meaningful reductions in opioid overprescribing are feasible and sustainable with central coordination and clinical leadership. The tools described in this report are accessible to most health systems. The drive to standardization for postsurgical prescribing has been facilitated with published reports of the degree of overprescribing and more appropriate postoperative starting doses.^37,38 We are aware that reducing a community’s opioid reservoir, may, in the short run, increase the number of persons who seek illicit street opioids and thus increase their risk of death. However, we regard cautious prescribing and opioid stewardship as a shared commitment to the long-term health of our community.

Accepted for Publication: July 9, 2018.

Published: September 28, 2018. doi:10.1001/jamanetworkopen.2018.2908

Open Access: This is an open access article distributed under the terms of the CC-BY License. © 2018 Meisenberg BR et al. JAMA Network Open.

Corresponding Author: Barry R. Meisenberg, MD, 2001 Medical Pkwy, Annapolis, MD 21401 (meisenberg@aahs.org).

Author Contributions: Dr Meisenberg had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Concept and design: Meisenberg, Grover, Campbell.

Acquisition, analysis, or interpretation of data: Meisenberg, Grover, Korpon.

Drafting of the manuscript: Meisenberg, Grover, Korpon.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: Grover, Campbell, Korpon.

Administrative, technical, or material support: Meisenberg, Grover.

Supervision: Meisenberg.

Conflict of Interest Disclosures: None reported.

Additional Contributions: Mitch Karpman, PhD, Anne Arundel Health System Research Institute, prepared the regression analyses. Charlotte Sundel, Center for Healthcare Improvement, Department of Medicine, helped prepare data displays. Neither were compensated for their assistance.