Clinical Variation Insight: Opioid Prescription Rates
- Louis Potok
On the heels of Hospital Medicine 2019, I am pleased to cover hospitalist opioid prescribing as the latest in Agathos’ insight blog series. Dr. Syed Raza (CHI St. Luke’s Health The Woodlands Hospital) recently presented a poster presentation at Hospital Medicine 2019 on his hospitalist team’s experience with the insight and how opioid prescription rates have steadily declined once Agathos insights were introduced.
Opioid abuse is a public health crisis in the United States. There are many causes, including regulatory gaps, misaligned corporate incentives, and long-term economic trends. But individual clinicians also have a role to play in ensuring that these powerful and often-useful medications are used judiciously, with an eye towards potential harms. In our work with hospitalists, we built on recent guidance from groups such as the CDC, Opioid Wisely, and Choosing Wisely to produce actionable metrics about opioid prescribing patterns.
What can hospitalists do to reduce the risk of opioid abuse? Most of the risk associated with opioids occurs outside of the hospital setting, where patients may develop long-term dependence or have the opportunity to overdose in an unmonitored setting. While reduced opioid use within the hospital may also be advisable in some cases, it is harder to evaluate whether those decisions are appropriate and less important than curbing unwarranted outpatient prescribing.
On the other hand, Agathos’ medical advisors suggested that it is safest for a single provider (usually the primary care physician) to manage a patient’s long-term therapy. So we quickly identified medication prescribing at discharge as a key opportunity for hospitalist contributions to opioid management.
USE CASE #1: Whether to discharge on opioids
We started by measuring the binary decision of whether or not to prescribe opioids at discharge. This was a simple rate: what percent of discharged patients received an opioid prescription at discharge? Patients were attributed to the hospitalist who wrote the discharge order and were counted in the numerator no matter who wrote the prescription for opioids.
While in many cases clinicians only feel ownership over prescriptions they ordered, since hospitalists own the discharge process, it is then appropriate for them to evaluate the medications prescribed. We found large variation among our users here: some had rates under 10%, while others were as high as 50%.
In some cases it may be appropriate for hospitalists to discharge patients with opioids, and at some sites hospitalists disproportionately aligned to facilities and services (e.g., orthopedics, oncology) where opioids are more clinically warranted. Yet even after normalizing for such, the wide span points to meaningful differences in clinical decision-making. We showed users a bar graph of all hospitalists’ rate, so they could see where they fell on the distribution.
USE CASE #2: Dosage of opioids when prescribed
Once a hospitalist decides to prescribe opioids, there are associated decisions to make about which medication, what dosage, and over what duration. Even in patients for whom prescribing opioids is appropriate, guidelines advise using the smallest effective dose for the shortest possible duration. This is a complex decision along several dimensions—how could we simplify this into a single, easy-to-understand metric?
We hit on the idea of calculating, for each prescription, the “total dose”. We start with the strength of a single dose. Different types of opioids have different strengths, which can be compared using their “Morphine Milligram Equivalent” (MME). Then, we multiply the MME per dose by the number of doses in the prescription and the number of refills. We add this up for all opioids prescribed at discharge per patient to get a single number which roughly quantifies the “intensity” of opioid prescriptions for that patient. While this is conceptually straightforward, it required some painstaking data cleaning to match medication types to their MMEs, and to parse out the many ways that dosages and refill quantities can be recorded across EHRs.
The attribution here is slightly different than used for the prescription rate metric. Here, we only include prescriptions written by the user, since these decisions are more fine-grained.
Finally, we capped the per-patient MME to 2000 (to limit the impact of outlier regimens, such as for oncology patients) and averaged over all patients where that clinician wrote any discharge prescriptions for opioids. Here too, we saw wide variation between hospitalists. Some of our users had an average “total dose” as low as 100 MMEs, while others were as high as 500. Here too, we showed a bar graph, as well as an auxiliary histogram breaking down of how many of the user’s prescriptions fell into different total dose “buckets”.
First, there was the clinical question of whether to include patients who were already on opioids when they entered the hospital. We chose to include those, upon guidance from our medical advisors that opioids should typically be managed by a single doctor (usually the primary care provider) and that withdrawal is uncomfortable but not dangerous (unlike, for example, withdrawal from benzodiazepines).
Second, we needed to design how best to explain and present our approach within the Agathos application. Phones are small, and physician time and attention are limited, so our general goal was to balance an ability to quickly understand the main thrust of the metric with an ability to drill down deeper into the methodology. We chose to link out to several external guidelines in addition to the explanatory text. Many states have specific laws or procedures around opioid prescription, so the application includes state-specific links and text, as per the hospital’s location. We chose to include tramadol, which is not technically an opioid but has a similar mechanism of action—so we included a link justifying that choice as well.
As with all insights, Agathos sends each hospitalist a personalized text message on their prescribing patterns for both opioid prescription rates at discharge, peer compare graphs, reference materials, and relevant cases within the Agathos mobile application. A subset of physicians also participated in small group sessions to review the data and discuss their practices. Examples of peer comparison graphs for the two use cases at a representative site are displayed below.
This was a particularly exciting insight for the Agathos team to develop, since it is mainly a quality play where the value reaches beyond the walls of the hospital. Results have further improved since Dr. Raza’s presentation, capturing the attention of academic and community thought leaders and publication interest.
Equally important, insights oriented around quality of care (rather than process adherence and cost) drive clinician engagement and buy-in. (Even some clinical insights such as lab utilization are rightly or wrongly perceived to be primarily about costs). While hospitalists’ value proposition is both quality improvement and cost reduction, the former can be more interesting and motivating for physicians who got into medicine for patient care. Since the practice economics of opioid prescribing at discharge are negligible from the hospitals’ vantage point, clinicians perceive feedback about practice patterns as squarely in the patient’s corner. This ultimately benefits hospitals looking to win clinician alignment.
As in this case, clinical best practices can change over time due to several factors, including advances in medical knowledge, changes in population health, and economic factors that change the costs and benefits of a given decision. The effectiveness of quality improvement and variation reduction initiatives hinges on a data-driven and user-centric approach:
- pull data from clinical records
- calculate physician metrics with a convincing attribution methodology
- prioritize findings among all other possible insights
- deliver feedback in an engaging, persuasive, respectful way
Curious about your group’s clinical variation? Contact us to request a demo.