Interventions to Reduce Over-Utilized Tests and Treatments in Bronchiolitis

The study was conducted at a free-standing, quaternary-care children’s hospital. Emergency, urgent, and inpatient care are provided at the main campus (∼500 beds) and 6 satellite locations totaling ∼19 000 inpatient admissions annually. Bronchiolitis patients are seen and admitted at all sites, totaling >2000 visits and ∼700 admissions per year with a median length of stay (LOS) of 60 hours. In the project interventions, we targeted all ED, UC, and inpatient sites. In September 2011, our hospital began using a CPG and an inpatient electronic health record (EHR) order set. These were the only previous formal bronchiolitis improvement initiatives at our institution.

Only patients who were seen during bronchiolitis season were included in the study. On the basis of bronchiolitis patient volumes during previous years, we defined bronchiolitis season as December 1 to April 30 of each year. For analysis, baseline data included 2 bronchiolitis seasons (December 1, 2013, to April 30, 2014, and December 1, 2014, to April 30, 2015), and the intervention season included December 1, 2015, to April 30, 2016. We used International Classification of Diseases billing codes to identify our target population, which consisted of all admissions of patients aged 1 to 23 months with a primary or secondary diagnosis of bronchiolitis who did not require any ICU services (Supplemental Information). ICU patients may warrant additional testing and treatment and are out of the scope of the AAP CPGs, so they were excluded. Of important significance, although the published use studies we used for benchmarking excluded patients with primary or secondary discharge diagnoses of asthma^5,6 or pneumonia,⁵ we did not exclude these patients from our sample to ensure that our analysis captured all the target interventions received by patients in our sample. Given the algorithms used by coders at our institution, bronchiolitis patients without a true diagnosis of asthma who were treated with bronchodilators during their hospitalizations may be assigned a primary or secondary diagnosis of asthma after discharge.

Our 40-person, multidisciplinary project team included hospitalists, ED and/or UC providers, bedside nurses, respiratory therapists, clinical application specialists, pharmacists, and process-improvement specialists.

This project was reviewed and approved by the Children’s Hospital Organizational Research Risk and Quality Improvement Review Panel.

A multidisciplinary kickoff meeting was conducted to understand the problem and develop interventions (Fig 1). The project team reviewed data from previous seasons and compared institutional performances to national benchmarks.^4,–6,15 Next, team members worked in small groups to brainstorm problem statements, key drivers that were unique to each resource, project aims, and proposed interventions. Finally, after a large group consensus, working groups formed to develop and implement each of the interventions detailed below.

Showing clinicians their data on performance compared with peers has been shown to be an effective and sustainable method of improving guideline-concordant care.^16,–19 For this project, each provider received 2 e-mails comparing their individual CXR, RVT, and bronchodilator use rates to their unidentified peers (Supplemental Information). The first e-mail in January 2016 showed preintervention data from December 2014 to April 2015, and the second e-mail in March 2016 showed intervention season data (December 2015 to February 2016). Each e-mail included inclusion and exclusion criteria and data limitations.

Efforts to reduce inappropriate antibiotic prescribing have shown that signing a pledge changes provider behavior and improves resource stewardship.^20,–22 Adopting this behavior-change method, we asked providers to sign a pledge to reduce unnecessary use in bronchiolitis (Supplemental Information).

The inclusion age changed from 1 to 15 months to 1 to 23 months to mirror the AAP guideline.³ The ED and inpatient management algorithms were moved to the first page, and visual cues (eg, a red stop sign for inclusion and exclusion criteria and yellow triangles for signs of deterioration) were added.²³ 

Inpatient EHR order sets were updated, and a link to the CPGs were added. The option for a trial of bronchodilators was removed, and an added alert reminded clinicians that CXRs, bronchodilators, and RVTs were not routinely indicated in bronchiolitis.

Education on the AAP guidelines and supporting evidence was provided at departmental and house staff meetings by using an audience response system to facilitate active audience participation. Institutional performance data, national benchmarks, project goals, and planned interventions were shared with audiences.

A patient-friendly handout and video were developed to educate and set family expectations early in their visits. A nursing order to perform education was included in the EHR order sets with embedded links to the handout and video.

Separate working groups were assembled for each intervention (eg, provider pledge and CPGs) on the basis of team member interest and expertise. The intervention working groups met weekly to monthly to plan and execute interventions. Interventions were implemented across all sites simultaneously. Informed by data and site champion input, the intervention working groups made changes to the interventions over time using rapid cycle improvement methods.

Inpatient and ED and/or UC site-specific champions for the main campus and 6 satellite locations informed intervention development and modification by sharing contextual factors that were unique to their sites with the intervention working groups. Working clinically at their respective sites afforded site champions in-depth knowledge of their working areas that was essential to the success of the interventions at those sites. For example, whereas education on bronchodilators primarily targeted respiratory therapists, at 1 site, registered nurses were responsible for bronchodilator administration. This unique situation resulted in education being uniquely tailored to registered nurses at that site.

In addition to using the data dashboard (described below), site-specific data were compiled and e-mailed to each site champion monthly. The site champions shared these data with their clinical teams using existing communication structures, such as daily huddles. The champions also posted the data in poster form to foster continued attention to each site’s progress toward goals.

The larger project team met monthly for 7 months. During monthly meetings, each intervention working group and site champion reported on their progress and barriers, and the whole team reviewed and discussed the overall project data. Implementation was aided by a project slogan: “REST is Best” (Supplemental Information).

The project team used a dynamic data dashboard for real-time data collection and analysis. The dashboard allowed all project team members, regardless of their QI training or background, to monitor general trends in use over time that were specific to their areas of interest. To populate the dashboard, EHR data were extracted nightly via an extract, transform, and load process into a database and subsequently compiled into a user-friendly dashboard (Supplemental Information). The final dashboard, which was built in Tableau version 8.1 (Tableau Software, Seattle, WA), was accessible to all members of the project team via the hospital Intranet. Line graphs displayed resource use as a percentage of the total number of bronchiolitis patients by month, and data were displayed in table format by season. Filters allowed individual project team members to stratify the data by clinical unit, date range, payer, and other important criteria. The data could be easily extracted into Microsoft Excel, allowing project team leaders with QI expertise to make and update control charts. The dashboard data were refreshed daily, enabling a timely evaluation of the effectiveness of interventions.

We assessed the proportion of patients who received CXRs, RVT, and bronchodilators during eligible admissions. Influenza A and B polymerase chain reaction (PCR) tests were not included in RVT given that influenza testing is recommended to guide antiviral treatment decisions for hospitalized patients.

We measured the percent of providers who signed the pledge by practice location and provider type.

To evaluate the possible unintended consequences of reduced use, we followed several measures. We measured the percent of patients who met study inclusion criteria and had a revisit to the ED and/or UC or admission for bronchiolitis within 7 days of an ED and/or UC visit or hospitalization for bronchiolitis. We excluded patients who required ICU care from our target population, but to evaluate for increases in care escalations accompanying decreased use, we measured the proportion of patients who required ICU-level care during their index hospitalizations as a proportion of all index bronchiolitis admissions and the proportion of bronchiolitis patients who required ICU-level care on readmission as a proportion of all readmitted bronchiolitis patients.

Shewhart P charts²⁴ were used to continuously evaluate outcome measures during the intervention phase. On each P chart, we plotted the outcome of interest on the vertical axis versus time on the horizontal axis. Each data point represents 14 days of data, and data points outside of bronchiolitis season were not included in the analysis. The control limits were based on the binomial distribution of proportions, as indicated when an outcome is not rare. Charts were annotated with each intervention. During the project, preintervention control limits were extended, and special cause variation was annotated on each chart by using established special cause rules.²⁴ When special cause was detected, a new centerline and control limits were calculated, and control charts were monitored for sustained improvement.

Characteristics and outcome measures of the baseline cohort were compared with the intervention cohort by using χ² or Fisher’s exact tests for categorical variables and a Wilcoxon rank test for continuous variables. Analysis was done in SAS 9.4 (SAS Institute, Inc, Cary, NC). All statistical tests were performed as 2-sided tests with a 0.05 level of significance.

Given the clinical challenge of differentiating bronchiolitis from asthma, we conducted a sensitivity analysis of bronchodilator use in children 12 to 23 months of age to understand the effect of including patients up to 24 months old.

Overall, 6659 patients were included in our study (4411 at baseline and 2248 postintervention). Table 1 shows population-level pre- and postintervention comparisons of those admitted. We found no significant difference in the LOS, but we saw a significant change in the all patients refined diagnosis related groups (APR-DRG) severity classification (P ≤ .001), and more patients were classified as observation status postintervention (P ≤ .001).

We saw a significant reduction in the ordering of all interventions for admitted patients, including the use of CXRs (39.3% to 27.2%; P ≤ .001), RVT (31.9% to 26.3%; P = .008), and any bronchodilators (34.2% to 21.5%; P ≤ .001; Table 2). When the mean and control limits were extended from the baseline, the P charts used to analyze the percent of admitted patients receiving CXRs, RVT, and bronchodilators had special cause variation indicating improvement (Fig 2). For patients who were discharged from the ED and/or UC, special cause variation indicated improvement in CXR and bronchodilator use (Supplemental Information). We did not meet our project goals for any of our outcomes. Admitted patients were more likely to receive interventions than patients who were discharged from the ED and/or UC. Most interventions were ordered in the ED and/or UC before admission.

There was more bronchodilator use in the 12- to 23-month-old group compared with the 1- to 12-month-old group, but the difference in proportion preintervention to postintervention was similar in both age groups (Table 3).

Overall, 57.9% of providers signed the pledge, including 53% of ED and/or UC providers, 76% of hospitalists, and 57% of pediatric resident physicians. By provider type, 34.4% of advanced-practice providers and 73.8% of attending physicians signed the pledge (Supplemental Information).

Overall, there was no change in the rate of revisits or readmissions after the intervention. During the intervention period, 8.1% (182 of 2248) of patients had an unexpected return to care within 7 days of their initial encounters, which was the same (8.1% [357 of 4411]) compared with postintervention (P = .997). There was no significant difference in the percent of patients with a hospital readmission within 7 days (1.7% [preintervention] and 1.0% [postintervention]; P = .21) for bronchiolitis. There was no change in the percent of admitted patients requiring ICU-level care during their index hospitalizations (23% [preintervention] and 21% [postintervention]; P = .30) or readmissions (5% [preintervention] and 7% [postintervention]; P = .29; Fig 3).

Although we did not reach our targets, this coordinated, multisite, multidisciplinary improvement initiative significantly reduced the use of CXRs, RVT, and bronchodilators in children admitted for bronchiolitis without unintended consequences, such as ED revisits, readmission, or escalation of care to the ICU. Our approach included tools and interventions not previously described in published bronchiolitis QI studies.¹² To our knowledge, this is the first published QI project in which researchers use a data dashboard for real-time, dynamic data sharing.

The data dashboard enabled timely data sharing across 6 sites, sparking competition among sites and sustaining interest in the project over time. Although we don’t have objective usage data, several project team members reported near-daily dashboard usage during the study. The provider pledge was a novel intervention that, to our knowledge, has never been used for a large, hospital-wide, QI initiative. The introduction of the pledge was associated temporally with special cause for inpatient bronchodilator use (Fig 2) and ED patient CXR use (Supplemental Information). However, the pledge was introduced simultaneously with staff education, so we cannot confidently conclude that the pledge was the intervention responsible for the improvement in these outcomes. The pledge was signed by most providers at our institution. The proportion of physicians who signed the pledge was higher than the proportion of advanced-practice providers. It is unclear how many providers consciously chose not to sign the pledge versus were not being given ample opportunity. We noted varying interpretations of the pledge with some, particularly trainees, finding it difficult to identify patients who required nonroutine care.

As has been shown in other disease processes, showing clinicians their performance data compared with peers likely positively influenced their behavior.^17,–19 The combination of signing a pledge to reduce use and seeing individualized data may have effectively influenced clinicians’ evaluations and management decisions. Our slogan, REST is Best, was quoted by clinicians throughout the institution, signaling that it was also an effective way of keeping our project goals visible.

Our results are of particular significance given that our institution’s preintervention CXR and bronchodilator use was already comparable to the best-performing quartile of tertiary-care children’s hospitals.⁶ Despite low baseline rates, we reduced CXR use to 27.2% of admitted patients, surpassing both the PHIS ABC (32.4%)⁵ and the ABC derived from a systematic review of previously published bronchiolitis QI projects (42%).¹² We also reduced bronchodilator use to 21.5%, approaching that of the PHIS ABC (18.9%).⁵ Given these results, the most likely reason we did not meet our project goals was that our goals were too ambitious.

Because of the inherent limitations in discharge billing codes, we included patients with pneumonia and asthma codiagnoses. This was both a strength and a potential limitation of our study. As intended, our broad inclusion criteria likely captured patients with bronchiolitis whom were unnecessarily treated with antibiotics or bronchodilators. Including these codiagnoses likely overestimated our use compared with the published benchmarks mentioned above, which excluded patients with these codiagnoses.^5,6,12 However, it is possible that differences in the proportion of patients with these codiagnoses in the pre-and postintervention periods may partially explain our findings. This potential limitation was mitigated by including all patients with bronchiolitis in our analysis rather than using traditional QI sampling techniques.

Our study has other important limitations. Given the seasonal nature of bronchiolitis, interventions were implemented on a rolling basis, so we are unable to determine with certainty which interventions changed provider behavior for which patients and through what mechanism, thereby limiting generalizability across settings and disease processes. We did not include a control group, so it is possible that our findings are the result of secular trends rather than directly related to our interventions. Finally, more patients were classified as minor APR-DRG severity and admitted to observation rather than inpatient status postintervention. This could indicate that patients in the postintervention period were less sick, contributing to the decrease in use we observed. However, given that our postintervention median LOS for inpatients was unchanged and there was no change in the percent of admitted patients requiring ICU-level care, it is more likely that fewer interventions in the postintervention period, together with payer influence, resulted in these changes.

Novel interventions significantly reduced overuse in bronchiolitis at our institution without unintended consequences for patients. Given that our baseline use was comparable to the best-performing quartile of tertiary-care children’s hospitals,⁶ our study has implications for other top-performing hospitals that are considering similar projects. Future researchers should explore provider attitudes, differences in acceptability by provider type, unintended consequences, and the effectiveness of the provider pledge. Finally, studies in which researchers use quasi-experimental methods are needed to determine which of our interventions were the most effective.

AAP

American Academy of Pediatrics

ABC

achievable benchmark of care

APR-DRG

all patients refined diagnosis related groups

CPG

clinical practice guideline

CXR

chest radiograph

ED

emergency department

EHR

electronic health record

LOS

length of stay

PCR

polymerase chain reaction

PHIS

Pediatric Health Information System

QI

quality improvement

RVT

respiratory viral testing

UC

urgent care

We thank Danielle Anderson; Joyce Baker, MBA, RRT-NPS, AE-C; Michelle Brown, MD; Stephanie Cannon; Cheryl Cavallaro, MSN, RN, CPNP; David Chung, MD; Beth Davis, BSN, RN, CPEN; Deb Dojka; Lauren Doty, BSN, RN; Ben Elkon, MD; Jason French, MD; Sarah Halsted, MD; Matthew Kopetsky, MS; Clint Hyatt; Megan Kirkley, MD, MPH; Marla Laufer, MD; Derrek Massanari, MD; Jackie McGee; Sonja Nickels; Danella Pochman; Janette Prokop; Casey Rabe; Lindsey Shaw; Laura Jean Turcotte, BHK, RRT; Misty Vivian; Kaitlin Widmer, MD; Lori Williamson, BA, RRT-NPS; and Jordan Wright, MD.