Impact of an Intervention to Reduce Bronchodilator Use in Bronchiolitis – A Time Series Analysis

Our institution is a 344-bed teaching and research institution. The ED sees over 90 000 patients per year, with approximately 1700 infants presenting with bronchiolitis annually. Hospital staff author and publish evidence-based clinical practice guidelines, with the guideline for bronchiolitis being updated every 3 years. Since May 2017, the guidelines have stated “medications are not indicated in the treatment of bronchiolitis.”

Interventions to reduce rates of bronchodilator prescription were codesigned with clinicians, EMR experts, a project officer, and medical lead (J.L.). Interventions were chosen based on a review of the evidence, ability to address identified drivers, and assessment of the feasibility within our context.¹⁰ 

Interventions were introduced on February 4, 2019 and included:

• Hour-long interactive education sessions conducted by the medical lead (J.L.) and delivered at ED staff education meetings, General Pediatric departmental teaching, and nursing education sessions over February 2019.

• Monthly audit-feedback tool via e-mail to Head of Department of General Medicine and ED to disseminate to frontline clinicians providing visibility of departmental rates of LVC practices in bronchiolitis – until February 2020 (Fig 1).

• BPA in the EMR for infants ordered bronchodilators in the ED (Fig 2) – this alert showed at the time of ordering if an ED physician entered an order for bronchodilators (salbutamol, terbutaline) for any child under the age of 1 year, regardless of diagnosis. The BPA offered educational advice: “Bronchodilators, including salbutamol, are not recommended for patients under the age of 1.” The BPA was automatically queued to “remove” the order, but this could be amended to “keep” if the clinician wished to continue with prescription, at which point the clinician was asked to enter a justification (acknowledge reason) - introduced February fourth 2019.

We extracted EMR data between May 1, 2016 and August 31, 2022 for all children under the age of 1 year given a diagnosis of bronchiolitis in the ED, excluding infants admitted to intensive care units as the level of severity may warrant different approaches to treatment. Data extracted included child demographics (age, sex), time of presentation, bronchodilator prescription details (time, login department of the clinician, and location of the patient at the time of order), length of stay, and admission ward. These data were used to create monthly aggregated rates of bronchodilator prescribing for ED presentations (rate of bronchodilator prescribing by doctors using an ED log in among all patients with bronchiolitis presenting to the ED).

BPA data were extracted from an EMR report that collected how many times the alert fired, ordering clinician details and action taken (including reason cited for over-riding the alert if relevant).

An interrupted time series analysis was performed on aggregated monthly count data for bronchodilator orders, offset by the size of the total population presenting each month with bronchiolitis between May 2016 and August 2022 (76 monthly intervals). An intervention date of February 4, 2019 was used as the date the BPA became live in the system.

Poisson regression analysis was performed to test the impact of the intervention over time: The model included: a constant, a baseline slope to control for secular trends, and terms estimating changes in the slope of outcome rates. The following equation was used: Yt = B0+B1T+B2Xt+B3TXt where B0 represents the baseline level at the study start, B1 is the change in outcome each time unit increase (preintervention trend), B2 is the level change (step effect), and B3 is the slope change following the intervention.²¹  Because of the nature of our intervention, including education and audit-feedback tools, we expected a long-term gradual impact, best captured by a change in slope, thus step-change was not included in our testing. Sensitivity analysis was performed as part of robustness checks mainly to allow for autocorrelation and seasonality.²¹ 

Analyses were conducted using Stata statistical software version 16 (Texas, USA).

Ethics approval for quality improvement project was obtained from the [names redacted for blinding purposes].

A total of 4635 infants preintervention (May 1, 2016–February 3, 2019) and 4948 infants postintervention (February 4, 2019–August 31, 2022) were diagnosed with bronchiolitis in the ED (Table 1). There was no difference in mean age, sex, or length of stay between the 2 cohorts.

Bronchodilator orders by ED staff were reduced from 6.89% (n = 319) to 3.23% (n = 160).

Males were more likely than females to be ordered a bronchodilator (8% vs 5.5%, P < .0001). Age was positively associated with bronchodilator prescribing, with the average age of a child ordered a bronchodilator 9.4 months compared with 5 months (P < .0001). Children admitted to the ward had much higher rates of bronchodilator use than children who were not admitted, indicating a positive correlation between perceived severity and bronchodilator use (12.7% vs 4.3%).

Controlling for baseline trend, the rate of bronchodilator prescription showed a decline of 2.7% following the intervention (P < .001). The results are robust to other specifications of the model, allowing for overdispersion and seasonality (Table 2, Fig 3). Figure 3 demonstrates the seasonal nature of prescribing in response to the seasonal nature of bronchiolitis, with higher presentation rates over winter. The deseasonalized trend line follows the seasonal trend, indicating the association is largely unaffected even after deseasonalizing the data (respiratory rate 0.98 95% confidence interval [0.96–0.99]). The slight deflection in the slope at the date of intervention also visualizes the change in slope effect.

BPA data: The BPA activated for 200 unique patients representing 4% of the postintervention bronchiolitis cohort (n = 4948). The alert fired between 1 and 12 times (median = 2) per patient. One hundred and eighty (90%) patients were ultimately prescribed salbutamol by a clinician and 10% avoided prescription. Reasons for over-riding the alert were themed with the most common reasons cited “almost 12 months old” (n = 98), “previous response” (n = 34) or “family history of asthma” (n = 22).

Following the implementation of a multifaceted intervention in the ED aimed at reducing rates of bronchodilator prescription in bronchiolitis, we found a statistically significant acceleration in the reduction of prescribing over time. The clinical impact of this sustained improvement, with fewer children exposed to unnecessary medication, will ultimately reduce the risk of side effects, clinician workload, environmental harm, and cost.

Previous studies looking to reduce LVC in bronchiolitis have used pre and post data.^11,15  Studies reporting on pre and post statistics may in fact simply capture time-based trends, falsely concluding that the intervention played a role. Our robust study design overcomes this by using large EMR datasets over 6 years to establish clear underlying trends.

The tighter definition of bronchiolitis used in Australia includes only children under the age of 1 year.^5,6  This may account for our favorable rates of bronchodilator prescribing compared with rates reported internationally.^9,12–17  This is also likely to reduce the risk of reclassification following bronchodilator prescription as asthma is rarely prescribed in our context under the age of 1 year.

Although our pre and post rates of bronchodilator prescription by ED physicians showed impressive improvement over time with reduction in prescribing rate from 6.9% to 3.2% post intervention, our analysis allowed us to better isolate the contribution of the intervention over and above pre-existing prescribing rates, and we were able to demonstrate a further 3% change in slope following the intervention (P = .04). This finding supports the role for BPAs as a powerful tool for behavior change in a more meaningful way than previous studies.

Although our intervention was multifaceted, the ongoing impact sustained over a 3-year period suggests that the most effective component of the intervention was the EMR alert. Education has a transient impact, which is unlikely to be sustained beyond the first rotation of junior staff (3 months), and the audit-feedback tool was ceased after a year because of resource limitations.

Over the past decade, with the widespread adoption of EMRs, observational and experimental studies have reported success in using EMR based interventions to reduce low value testing. Most published reports use BPAs, providing clinicians with decision support at the time of ordering for laboratory tests and imaging. A recent systematic review looked at 122 trials involving EMR interventions designed to improve clinician adherence to best practice guidelines.²²  This review showed a modest absolute improvement in percentage of patients receiving the targeted process (5.8%) and a slightly lower average rate of improvement (4.4%) for the 68 trials looking specifically at prescribing behavior. Most of these studies used pre and post comparisons, making our study, with more robust methods, difficult to directly compare.

Significant heterogeneity between studies in the systematic review was observed with insufficient evidence to draw conclusions around which aspects of an EMR alert led to a more successful outcome. There is some evidence to suggest that a hard stop, where the user is prevented from progressing, is more effective than an alert that allows the user to bypass without any justification.²³  In a pre and post study of radiology orders, alerts that required action had a 10-fold higher rate of action.²⁴  Given our BPA supports a recommendation to follow evidence-based guidelines rather than a serious safety risk, a hard stop would likely lead to substantial clinician frustration and therefore was considered inappropriate. However, to optimize the effectiveness, we asked users to justify their decision if they chose to proceed with the prescription. This also allowed us to collect data over time regarding ongoing drivers of prescribing behavior. Understanding these drivers can help inform future education efforts.

Few studies consider alert fatigue, an important safety consideration.²²  Frequently over-ridden alerts can contribute to clinician frustration and a tendency to ignore all alerts, even those that present critical safety information.²⁴  Although our results show an overall reduction in bronchodilator prescribing, reviewing the response to the alert is important to understand the contribution to alert fatigue. Our data show the alert fired up to 12 times per patient. This suggests that the alert may have been effective for some clinicians and not others, or that a single prescriber activated the alert several times before making a final decision whether to prescribe. Although the ultimate outcome was to prescribe in 90% of cases, the interruption in workflow and educational messaging may have had an impact on the clinician’s decision making in future cases. Potentially, at the time of the BPA firing, the conversation had already been had with the family and it was challenging for the clinician to return and offer a different management plan. Understanding the reasons for prescribing allows future educational campaigns to target misconceptions of age, family history, and perceived prior response.

There is no current agreement on what constitutes an acceptable rate of over-riding BPAs and it is likely to be context-specific depending on the risks. In-depth interviews with prescribers would provide useful insights to inform future iterations.

The greatest strength of our study is the use of a robust analysis that uses EMR data and accounts for pre-existing prescribing trends. Our ability to measure change in ordering practices over 6 years allow us to account for the underlying trend toward improvement and isolate the additional impact of the intervention.

Secondly, we use codesigned interventions that are more likely to be acceptable to end-users. By involving ED clinicians in the design of the BPA, we were able to optimize the design and minimize the risk of clinician frustration with increasing alerts in the EMR. We also captured rates of alert response and reasons for prescribing, which will help inform future efforts to address LVC, both in bronchiolitis, but also for other pediatric LVC practices.

Our study has some limitations. We can only report on association, not causation. We assume that the most effective aspect of our intervention was the EMR alert, given the sustained impact over time, but can only measure the impact of the intervention as a whole. We tested our intervention in 1 hospital and only in the ED. Given our institution is a large teaching hospital with strong clinical champions, the findings may not be generalizable to other institutions or beyond the ED. Potentially the BPA would play a larger role in institutions without the strong pre-existing trend toward deimplementation.

An EMR alert, as part of a multifaceted intervention to reduce the use of bronchodilators in bronchiolitis, appears effective. The findings are consistent with other LVC studies in bronchiolitis that suggest system-level changes as being more impactful than individual clinician interventions. The impact must be balanced against the risk of alert fatigue and our BPA design allowed us to collect ongoing information about drivers and rates of alert acceptance. Compared with education and audit-feedback tools, EMR interventions require less ongoing resource likely representing a high value intervention. Robust evaluation of interventions to reduce LVC is critical to ensure investment in ineffective interventions does not add unnecessary cost to our burdened healthcare systems. Further research to understand the impact of EMR interventions alone, and the cost-effectiveness in LVC, is warranted.