Choice Architecture in Opioid Safety Alerting

Abstract

The need for effective and efficient clinical decision support (CDS) embedded in electronic health record (EHR) processes is growing. Using choice architecture design strategies may increase effectiveness of CDS solutions. The authors describe implementation of an opioid risk alert and subsequent revisions of that alert to increase effectiveness and reduce alert volumes. The first version of the alert used an opt-in choice architecture when recommending naloxone and the second version used an active choice design. The percentage of opioid prescriptions ordered with naloxone prescribed within the last 12 months increased significantly after implementation of the first version of the alert and then further increased significantly after implementation of the second version. Alert volumes decreased over the same timeframe. An education campaign was also implemented during the timeframe studied and likely also contributed to the naloxone outcomes seen.

Introduction

Electronic health records (EHR) have become near ubiquitous since the adoption of the Health Information Technology for Economic and Clinical Health act. These systems have created many opportunities to inform decisions about patient care. Clinical decision supports (CDS) are integrated into the EHR to provide timely information at the point of care that can lead to higher quality outcomes.

While EHR embedded alerts are commonly used to deliver CDS, evidence suggests as the number of alerts in a system increases, overall alert effectiveness decreases¹. Additionally, many alerts are interruptive, which may lead to medical errors including failure to return to interrupted tasks^2,3. To reduce the effects of alert fatigue, organizations may focus on reducing alert volumes⁴. Organizations may also consider non-interruptive CDS strategies, which in some cases may be more effective than interruptive alerts⁵. In this paper, we describe our approach to increase the effectiveness of an opioid risk alert by changing its choice architecture.

Choice Architecture

Applying behavioral economics principles to decision support design has been shown to improve effectiveness of technical solutions intended to influence human behavior^{6, 7, 8, 9}. These studies show that the way a choice is presented to a person can have a large effect on the selection made. This type of choice architecture commonly takes on one of two forms: defaults or active choice.

Defaults may be designed with an opt in or an opt out approach. In either case, if the person takes no action, the default will stand. Active choice requires a person to choose between two or more options before proceeding. In this case, a choice is required and there is no default selection.

The best choice architecture approach for any given scenario will depend on various factors. The use of opt-out approaches, where an option is selected by default, should be carefully considered. There is risk of the default value being inappropriately ordered¹⁰. When an option is expected to be in the best interest in most cases but may also cause harm in others, an active choice may be best. Opt-in approaches, where an option is made available but not selected by default, may be most appropriate in cases where there is less certainty about which option will be best.

Opioid Epidemic

There has been a rapid increase in opioid related overdose deaths in the United States in recent years. More than half a million people have died from prescription and illicit opioid related overdose since 1999¹¹. This crisis has been appropriately referred to as the opioid epidemic. Healthcare systems across the nation are now focused on increasing opioid safety.

Through system standardization and caregiver engagement, Intermountain Health has increased community awareness of the risks associated with prescribing and taking opioids. Several factors affect a person’s risk of dying from an opioid overdose. Risk of overdose death increases as the potency of the opioid dose taken increases¹². Likewise, taking benzodiazepines, a class of sedatives, with opioids increases risk of overdose death¹².

Unfortunately, even though opioid prescribing in the U.S. has decreased substantially over the past several years, opioid overdose deaths have increased 3-fold due to increased use of illicit opioids¹³. This alarming information prompted system leaders at Intermountain Health to more strongly focus on increasing access to naloxone, a medication that can save the life of a person suffering an opioid overdose, whether caused by prescription or illicit substances.

The FDA recommends healthcare professionals discuss the importance of naloxone with all patients receiving prescription opioids to treat pain and consider prescribing naloxone for any patient at increased risk of opioid overdose even if they are not receiving prescription opioids¹⁴.

Paper Objective

In this paper, the authors describe technical and non-technical interventions deployed to increase opioid safety at Intermountain Health. The technical solution implemented, a CDS alert, underwent major revisions in October 2022 to better align with the choice architecture principles described above. The effect of the original and revised alert on key outcomes of interest is described in detail.

Methods

In 2020, a system-wide goal at Intermountain Health was set to reduce the proportion of opioid prescriptions ordered at greater than 90 morphine milligram equivalents (MME) per day. In 2022, system leaders set a goal to increase naloxone prescribing across the system for community safety. Multiple interventions were deployed to support these goals, including education and communication, and CDS implemented in the EHR. It is important to note the number of ordered prescriptions that are filled is unknown.

Education and Communication

In 2020 and 2021 an opioid safety education and communication campaign were launched helping to reduce stigma regarding opioid prescribing best practices. Dashboards were created to provide transparency to service lines with their data regarding high MME opioid prescribing patterns. A website with opioid resource information was created, and letters were sent to prescribers to communicate the goal to reduce opioid MME potency if and where appropriate. Rounding with healthcare professionals, webinars, and additional meetings were facilitated to bring awareness to the opioid safety initiative. A Questions and Answers video interview was also developed for prescribers to understand complex questions surrounding opioid safety. Multiple avenues of communication were sent out to the system including banners, screensavers, and posters relaying the importance of opioid safety. Individual education sessions were also conducted with system top prescribers of opioids.

In 2022, an education and communication campaign were undertaken to spread awareness of the system goal to increase transparency of the importance of naloxone prescribing. Multiple teams collaborated to build education materials for streamlining communication to prescribers. These materials included naloxone conversation starters, flyers for naloxone importance, and steps to take with naloxone for opioid overdose. The analogy of prescribing naloxone was compared to having a fire extinguisher at home, one would hope it’s not needed, but should have on hand in case of emergency. Prescriber education was communicated to all areas of the system with additional focus to areas with high opioid prescribers and minimal naloxone prescribing.

In addition to focusing on educating prescribers, multiple news segments, podcasts, and articles regarding opioid safety and importance of naloxone were communicated through an external campaign to the community. System leaders also attended the Utah Pharmacy Association convention where they discussed new prescribing laws in Utah for pharmacists and the role they play in naloxone awareness.

EHR Embedded CDS

In August 2020 an opioid safety alert (V1 Alert) containing several components to assist prescribers in best opioid prescribing practice was implemented in Cerner, Intermountain Health’s core EHR system. The original design of the alert was provided by the EHR vendor. Adjustments were made to the alert wording and three of the alerting criteria thought likely to be less helpful in decision making best practices were excluded before implementation. In November 2020 an opioid induced respiratory depression risk index¹⁵ was added to the alert to increase patient safety. This was in response to Intermountain Health recognizing the need to bring awareness to and assist prescribers in assessing the magnitude of an individual’s potential risk for opioid induced respiratory depression.

The V1 Alert message described the opioid overdose risk factors identified and recommended naloxone. The V1 Alert did not require the healthcare professional to actively choose between prescribing vs not prescribing naloxone. Instead, it provided an option to order naloxone and the option was unselected by default.

In December 2021 the V1 Alert was modified to reduce nuisance alerting. The alert was suppressed for individual prescribers on individual patient charts for 60 minutes after the initial alert displayed. This reduced duplicate alerting in cases where initial and subsequent future-date prescriptions were written at the same time. Additionally, the opioid prescribing alert MME calculation was changed to only consider single incoming prescriptions and the alert did not add up MME values across prescriptions. This helped prevent the possibility of over-estimation of MME values in cases where a prescription was renewed before its end date had been reached. In July 2022 the V1 Alert was modified to remove alerting for most non-prescriber roles where medication treatment decision making would not be applicable.

To promote increased naloxone prescribing, a new version of the opioid safety alert (V2 Alert) replaced the V1 Alert in October 2022. The V2 Alert retained the same alerting criteria except the MME threshold was decreased from 90 MME down to 50 MME. The V2 Alert was also changed to not alert if naloxone had already been prescribed within the last 12 months (Table 1).

Table 1.

Alert Elements by Version.

The V2 Alert message was greatly simplified compared to the V1 Alert to increase ease of understanding. The recommendation to order naloxone was more highly emphasized in the V2 Alert and the choice architecture was changed from an opt in approach as described above, to an active choice design (Figure 1). After hearing feedback from prescribers that the V2 Alert was displaying in some cases where a patient was not prescribed an opioid, the error was identified and fixed in the V2 Alert logic in December 2022.

Figure 1.

V1 and V2 Alert Choice Architecture Design.

Results

Two key clinical outcomes were measured. The proportion of opioid prescriptions ordered where naloxone was prescribed within the last 12 months is shown in Figure 2. The proportion of opioid prescriptions ordered with a dosing regimen at greater than or equal to 90 MME per day is shown in Figure 3. Pre/post measures of statistical significance were calculated using a 2X2 Pearson’s Chi-Squared test.

Figure 2.

Opioid Orders with Naloxone Prescribed within the Last 12 Months.

Figure 3.

Total and % of Opioid Orders at >= 90 MME Per Day.

After implementation of the V1 Alert, the mean percentage of opioid prescriptions ordered with naloxone prescribed in the last 12 months increased significantly (3.23% pre, 13.49% post, P < 0.0001). Likewise, the mean percentage of opioid prescriptions ordered at greater than or equal to 90 MME per day decreased significantly (9.81% pre, 7.09% post, P < .0001).

After implementation of the V2 Alert, the mean percentage of opioid prescriptions ordered with naloxone prescribed in the last 12 months increased again (12.53% pre, 20.59% post, P < .0001). The mean percentage of opioid prescriptions ordered at greater than or equal to 90 MME per day did not significantly change after implementation of the V2 Alert (5.45% pre, 5.59% post, P = .1365).

It is noted that the percentage of opioid prescriptions with naloxone prescribed in the last 12 months slowly increased over several months in mid to late 2022 and then increased more dramatically after implementation of the V2 Alert in October 2022. The earlier increase is likely attributable to system-wide education efforts implemented during that timeframe.

Alert volumes over time were measured (Figure 4). Annotation A in the figure identifies implementation of the V1 Alert. An increase in alerting is noted after introduction of the opioid induced respiratory depression risk score criteria in November 2020 (annotation B). A decrease in alerting is noted after changes were made in December 2021 to reduce nuisance alerts due to over-estimation of the MME value in cases of multiple prescriptions and prescription renewals (annotation C). Another decrease in alerting was seen after changes were made to remove alerting for most non-prescriber positions in July 2022 (annotation D). An initial increase followed by a substantial decrease in alerting is noted after implementation of the V2 Alert in October 2022 (annotation E). Shortly after implementation of the V2 Alert, reports of inappropriate alerting on benzodiazepine medications were reported. The issue was fixed in December 2022 (annotation F). Ongoing decreased alerting since introduction of the V2 Alert is most likely due to the suppression logic introduced to no longer alert for cases where naloxone has already been prescribed in the last 12 months.

Figure 4.

Alert Volumes. A) V1 Alert implemented, B) added opioid induced respiratory depression risk score, C) reduced over-estimation of MME, D) removed alerting for most non-prescribers, E) V2 Alert implemented, F) corrected error in V2 Alert logic.

Discussion and Conclusions

In this paper the authors report improved clinical outcomes and reduced alert volumes with ongoing revisions made to an opioid risk alert. One major revision made was changing from an opt-in choice architecture to an active choice design associated with a recommendation to order the life-saving medication naloxone. The percentage of opioid prescriptions ordered with naloxone increased after implementation of the opt-in design and then increased further after switching to an active choice design. It is unknown whether the percentage of opioid prescriptions ordered with naloxone would have increased more rapidly if the active choice design had been implemented in the first version of the alert.

The ultimate objective of the CDS described in this paper is to reduce opioid overdose deaths in Utah. Fortunately, Utah has seen a reduction in the proportion of overdose deaths due to prescription opioids over recent years. The proportion of overdose deaths due to the illicit opioid heroin is also decreasing. However, the proportion of overdose deaths due to the illicit opioid fentanyl is increasing. Additionally, the proportion of overdose deaths due to methamphetamines, which are not opioids, is greatly increasing (Figure 5). While total overdose deaths in Utah consistently decreased each year from 2015 to 2019, deaths increased from 2019 to 2020 (Figure 6). More recent total overdose death data was not available to the authors at the time of this publication.

Figure 5.

Overdose Deaths in Utah. Source: Utah Department of Health and Human Services Office of the Medical Examiner. 2022. “Fatal Drug Overdose Data Update.” Internal Report Provided to the Authors Upon Request.

Figure 6.

Utah Overdose Deaths. Source: http://wonder.cdc.gov

It is important to note while having naloxone on hand is highly recommended for those at risk of opioid overdose, this life-saving medication is not fully covered by all payors and the out-of-pocket cost may be an access barrier for many patients. Additionally, prior authorization requirements can add a significant burden for naloxone prescribers. While the authors have had some success working with one of the local payors in their area, to be most effective, removal of these barriers should be addressed at a national level.

Limitations

Multiple interventions, both technical and non-technical, were implemented at various times throughout the course of the quality improvement work described in this paper. This makes it difficult to attribute changes in outcome variables to any single intervention.

Prescribing data was limited to orders placed in Intermountain Health’s core EHR used in Utah and parts of Idaho. Paper prescriptions and prescriptions ordered in other EHR’s were not included. Additionally, data about whether a prescription was filled at a pharmacy was not available and is not included in this report.

The V2 Alert, which replaced the opt-in choice architecture of the V1 Alert with an active choice design, was implemented only a few months before this paper was authored. More time to evaluate the outcomes of interest may reveal additional insights.

Conclusion

Given the growing need for effective CDS and significant risk of alert fatigue, intentional design of EHR based tools is imperative. As shown in this paper, applying principles of behavioral economics and choice architecture to decision support design may increase effectiveness and reduce alert burden. Likewise, continued monitoring and revision of decision support solutions over time are requisite to ensure movement toward more fully optimized CDS.