Pragmatic Expansion of a Neonatal Antibiotic Stewardship Program in a Community Health Care System Free

PAR is a community medical center in Athens, GA with the only Level 3 NICU in a 17-county service area. In 2016, the NICU was staffed by 2 neonatologists, 2 neonatal nurse practitioners, and 42 nurses, with an average daily census of 15 infants. The NBN was a 28-bed unit staffed by a team of 45 nurses and 3 pediatric hospitalists.

Piedmont Healthcare (PHC) is a rapidly expanding health care system in Georgia. In 2018, PHC included 10 delivery hospitals; all 10 had NBNs and 8 provided NICU care at level 2 or higher. In 2016, PAR (Suburban 1) joined the Vermont Oxford Network’s internet-based Newborn Improvement Collaborative for Quality (iNICQ) entitled “Choosing Antibiotics Wisely” conducted from January 2016 to December 2018.^7,8  Other PHC hospitals with NICUs were invited to join a Piedmont QI collaboration through iNICQ in 2018. All joined, including 1 that participated in iNICQ independently in 2017 (Suburban 2) and 6 others that had not yet undertaken efforts in this area. In one of the participating centers, the NBN was not included because all antibiotics were given in the NICU; in all others, both the NICU and NBN participated. The collaborative included 8 hospitals, with 8 NICUs and 7 NBNs participating. Because of limited AS resources, the 2 hospitals with only NBNs were not included.

Individual hospitals are described in Table 1. Most NICUs operate independently, except for 2 practices that staff multiple hospitals. Before the start of this project, there was no platform for communication, discussion, or practice change across NICUs within the system. This Piedmont QI collaboration includes 21 physicians, 16 neonatal nurse practitioners (NNPS), 410 nurses, 2 physician assistants (PAs), and 9 pharmacists across the state of Georgia.

The QI team at PAR consisted of a pediatric infectious disease specialist, neonatologists, a pediatric clinical pharmacist, neonatal nurses, and IT support. All team members were encouraged to view webinars provided by iNICQ with project updates immediately after the webinars. Strategies for improvement included SMART (specific, measurable, achievable, relevant, time-based) aims, plan-do-study-act cycles, antibiotic audits, manual data collection, data reporting, and standardization of practices through written guidelines and changes in the electronic health record (EHR). Interventions included a shortened sepsis rule-out period,^15–17  increased blood culture volume, the alteration of gentamicin dosing,¹⁸  a shortened duration of therapy, the alteration of clinical guidelines, and a parent education handout (Table 2).^17,19,20  To evaluate the benefit and feasibility of the implementation of the Kaiser neonatal sepsis risk calculator (SRC),²¹  the SRC was applied retrospectively to infants who received antibiotics in 2016. This was followed by an SRC pilot and then formal SRC implementation.

A presentation to PHC outlining a systemwide project reviewed the PAR successes, highlighting 180 fewer doses of ampicillin in the NICU, 22 fewer infants started on antibiotics in the NBN, 140 fewer doses of ampicillin and 82 fewer doses of gentamicin in the NBN, and an estimated reduction in hospital stay of 25 days in 2017 compared with 2016. It included the rationale for the project, benefits of iNICQ (eg, continuing medical education [CME] and maintenance of certification quality improvement credit [MOC4]), cost per institution, and a list of other participating Georgia centers. Based on this presentation, PHC executives approved the project. Subsequently, NICU directors from the 6 individual NICUs not previously involved in iNICQ were contacted via e-mail and phone and invited to join. The presentation was then shared with these NICU directors via e-mail. Each center was required to provide local funding. In July 2018, the presentation was shared with providers from one of the hospitals with only an NBN (Rural 1) via telephone and e-mail to assess interest in a phase 2 project. The tenth hospital (Urban 7) did not receive information about the project because of the lack of resources of the AS team.

All 8 NICUs joined iNICQ, with each center identifying a lead provider. AS pharmacists involved in the care of adult patients from each hospital were invited. Baseline antibiotic usage data were provided by the Duke Antimicrobial Stewardship Network, which was supporting adult AS at PHC. All providers were able to see the ampicillin and gentamicin antibiotic utilization rate (AUR) for their NICU compared with other NICUs. A system SMART aim was set: a 10% reduction in antimicrobial utilization among newborns (NICU and NBN) at each PHC hospital that takes care of newborns by December 2018.

A key driver diagram identified 5 primary drivers, with multiple secondary drivers and change ideas linked to each (Fig 1). An AS charter template was provided that included a project description, a statement of need, project aims, change ideas, potential measures, milestones, potential barriers to success, resources needed, communication plan, stakeholders, and team roles and responsibilities. Six of 8 hospitals modified the document to create hospital-specific plans. Two change ideas were agreed on across the system: 36-hour sepsis rule-out and single-dose gentamicin for 36-hour rule-out.^15,18  Five hospitals agreed to increase the focus on antibiotic education with families,^17,19,20  3 by incorporating education into rounds, and 2 by using paper handouts. All 8 NICU hospitals agreed on the incorporation of the SRC into the common EHR for potential use in the future. Three hospitals agreed to pilot the SRC with 3 more hospitals agreeing to incorporate it into the workflow after it was incorporated into the EHR. Three hospitals identified additional AS change ideas (Table 3).

AS education was provided via iNICQ webinars. Each webinar was followed by a PHC system video conference. System- and hospital-specific AURs, goals, and change ideas were reviewed at each meeting along with updates on the local rollout of changes across the system (Table 4). The involvement of all the NICUs in the health care system facilitated the prioritization of IT support to create a Tableau²²  dashboard enabling automated data collection, as well as the integration of the SRC into the health care system’s EHR.

In May 2019, all the NICU directors met in person for the first time. Updated hospital-specific antibiotic utilization data from Tableau were shared. AUR comparisons between hospitals were visible to all providers. A revised key driver diagram was shared, including an emphasis on SRC implementation and information on how the SRC could be accessed from the EHR. After the incorporation of SRC in the EHR, hospitals Urban 1 to 3 incorporated this tool into their workflow. In December 2019, 5 hospitals completed a team-based review of the web-based microlesson modules provided by iNICQ.

From January 2020 to July 2021, electronic data collection continued. During this time, there was no further central communication and no further AS initiatives.

The outcome measure was AUR in the NICU and NBN measured as antibiotic days per 1000 patient days. Each antibiotic prescribed in a day was counted as a separate antibiotic day. Automated data collection allowed for the retrospective calculation of antibiotic courses initiated per patient admitted over time. A balancing measure of restarts of antibiotics at 18 to 72 hours after antibiotics were stopped per 1000 patient days among newborn patients was monitored.

From 2016 to 2018, retrospective and prospective data were manually collected by appointed physicians or pharmacists at the PAR NBN, PAR NICU, Suburban 2 NICU, and Urban 6 NICU. In 2018, the development of the Tableau dashboard allowed the process to be done electronically from the common EHR. Retrospective data starting from January 2017 to June 2018 and prospective data from June 2018 to July 2021 was collected electronically from all hospitals by using the common EHR, including the 2 NBNs not involved. Statistical run charts combining manual and electronic data were used to evaluate the performance at individual hospitals starting in January 2017, except for the PAR NBN and PAR NICU, in which earlier manual data were incorporated. Data collection for the Urban 6 and 7 NBNs began after their joining the common EHR in October 2018. U-type control charts were used to evaluate performance across the system. Special cause variation for each chart type was determined by using rules suggested for health care.²³  A sustained decrease in AUR was defined as a 10% reduction in AUR at the end of the measurement period compared with the baseline AUR.

Institutional review board review was done by PAR initially and then by the PHC review board, and the project was considered QI.

Performance for the primary outcome measure at PAR (Suburban 1) is shown in Figs 2 and 3. Despite fewer doses of ampicillin and gentamicin in the years 2016 and 2017, the PAR NICU AUR did not decrease until year 3 (2018). At this time, the AUR decreased from 400 to 215 antibiotic days per 1000 patient days (Fig 2), correlating with the formation of the system-wide stewardship collaborative and the approval of a shortened antibiotic duration for culture-negative sepsis. During the first year of the project (2016), PAR NBN saw a decrease in AUR from 75 to 30 antibiotic days per 1000 patient days. Special cause variation was correlated with PAR joining iNICQ in January 2016. In the third year of the project, decreased variability in the PAR NBN AUR was seen with the initiation of the SRC pilot. During the intervention period, 2016 to 2019, there was a 46% decrease in the PAR NICU AUR and an 83% decrease in the PAR NBN AUR.

Performance for the primary outcome measure for PHC is seen in Figs 4 and 5, with hospital-specific data presented in Figs 6 and 7. In the third year of the project and the first year for the network (2018), the PHC NICU AUR decreased from 408 to 345 antibiotic days per 1000 patient days (Fig 4). A second decrease from 345 to 298 antibiotic days per 1000 patient days, correlated with 2 hospitals joining the common EHR. The PHC NBN AUR decreased from 59 to 42 antibiotic days per 1000 patient days (Fig 5), which correlated with the initiation of an SRC pilot. In the fourth year of the project and the second year for the network (2019), the PHC NICU AUR decreased from 298 to 246 antibiotic days per 1000 patient days, which correlated with the first in-person NICU directors’ meeting and the final hospital joining the common EHR. The NBN AUR decreased from 42 to 15 antibiotic days per 1000 patient days, which correlated with the incorporation of the SRC in the EHR. During the intervention period, 2018 to 2019, within PHC, there was a sustained 40% decrease in the NICU AUR and a 74% decrease in the NBN AUR. Seven of 8 NICUs showed sustained decreases in AUR (Figs 2 and 6). In 1 intervention hospital (Suburban 2), all antibiotics were given in the NICU, and thus, there was no involvement of the NBN. Five of 7 of the intervention NBNs showed sustained decreases in AUR (Figs 3 and 7). The NBN that was queried about possible involvement in phase 2 (Rural 1) had the highest baseline antibiotic use and saw a decrease in AUR after education about our project. Urban 6 did not receive any education about the project and did not see any decrease in antibiotic use.

Within PHC, the percentage of antibiotic courses initiated per patient admitted in the NICU decreased from 0.77 to 0.59, resulting in 335 fewer courses of antibiotics in 2020 compared with 2017. Within the NBN, antibiotic initiation decreased from 0.025 to 0.008, resulting in 189 fewer infants receiving antibiotics in 2020 compared with 2017.

In July 2020, there was a decrease in the balancing measure from 108 to 80 antibiotic restarts at 18 to 72 hours per 1000 patient days (Fig 8). During the 1.5-year post-observation period (Jan 2020 to July 2021), there was no rise in the PHC NICU or NBN AURs (Figs 4 and 5).

System-wide improvements led to a dramatic change in antibiotic usage among newborns with a sustained reduction in antibiotic use in either the NBN, NICU, or both in all intervention hospitals. The project was successful in engaging the support of all NICUs and many of the NBNs in the health care system.

The first decrease in PAR NBN AUR was noted after joining iNICQ. At that point, no formal interventions were initiated. Similarly, Rural 1 saw a decrease in AUR after education about the project, despite not being formally involved. This decrease in AUR was likely the result of conscious and subconscious changes in prescribing practices motivated by the knowledge that antibiotic utilization was going to be monitored and shared.

In the first 2 years of the project, PAR NICU tracked doses of ampicillin in early onset sepsis (EOS) as a process measure. Noting a decrease in doses of ampicillin encouraged the team to continue despite a lack of change in the AUR. There were no process measures tracked for the system-wide project because there were different interventions at different times in different hospitals.

A significant and sustained decrease in AUR was seen in 2 Piedmont hospitals (Suburban 1 and Suburban 2), with highly engaged teams that included pediatric-trained clinical pharmacists. Both centers were actively involved for at least 2 years and implemented multiple interventions before seeing a drop in AUR. These included the standardization of treatment guidelines, a shortened sepsis rule-out, a shorter duration of antibiotics for culture-negative sepsis, and the identification of infants at a low risk for sepsis. These results are like those described by Meyers, et al in the United States²⁴  and Maki, et al in the United Kingdom.²⁵ 

Tamma, et al²⁶  described the success of a large multimodal QI AS program involving 402 hospitals through the implementation of web-based and durable educational content, the external facilitation of trained QI and AS experts, and feedback on antibiotic use. Their 1-year program from 2017 to 2018 revealed a greater decrease in centers that were more involved in the program. Similarly, we found a correlation between the degree of involvement and the decrease in AUR.

The use of an “opt-in” strategy helped promote individual hospital engagement. Given the culture of independence within NICUs across the system, it was important for local leaders to choose to participate. A desire to improve care and the ability to receive MOC4 provided motivation for providers to participate.

Data sharing allowed providers to compare their hospital antibiotic use with other hospitals within the system. Like in the study by Meyers, et al,²⁴  this appeared to be a strong motivator for change, as evidenced by a decrease in the system AUR after initial data sharing and when data were shared at the in-person NICU directors meeting. Creating a partnership between previously independent NICUs validated the need for IT support to facilitate automated data collection and incorporate the SRC into the EHR. Both IT interventions made significant contributions to long-term sustainability.

Several hospital-specific factors likely influenced the changes in AUR. Urban hospitals 1 to 3 share the same physician pool, with physicians spending similar time at each clinical site. The variable AURs between these sites highlights the impact of the patient population (Urban 1 has a more affluent population with higher levels of prenatal care), as well as nursing culture (Urban 2 and 3 have similar patient populations but different nursing response to feedback). The Urban 4 NICU had the lowest baseline AUR with less room for improvement with small changes such as 36-hour rule-out and gentamicin dosing. The sustained drop in the NICU AUR at Urban 6 correlates with the transition to the common EHR, likely related to the standardization of care within the unit through built-in order sets. Rural 1 NBN baseline AUR was the highest, which was possibly related to a predominance of family medicine-trained physicians with less pediatric experience. These physicians were motivated to change, showing a dramatic drop in AUR with minimal intervention (1 phone call and 1 e-mailed presentation).

Local factors impacted how different centers approached the same aim: provider staffing models, nursing culture, and location or unit in which infants receive antibiotics. Urban hospitals 1 to 3, staffed by a common provider team, opted to incorporate antibiotic education and interventions into rounds, which seemed simpler than involving and training the 150 nurses among all 3 hospitals. Suburban 1 and Urban 5 both had small numbers of providers who worked exclusively within those units. Having fewer numbers of providers and nurses made it easier to run SRC pilots. At Suburban 2, all infants receiving antibiotics were admitted to the NICU; this well-organized QI team was able to implement an SRC pilot and a hard stop for the 36-hour sepsis rule-out.

The balancing measure, antibiotic restarts at 18 to 72 hours after antibiotics were stopped per 1000 patient days, decreased in July 2020. This was 1.5 years after seeing a decrease in AUR and likely correlates with increased physician confidence with the practice changes associated with AS.

We were unable to fully engage 2 hospitals and did not see a sustained drop in antibiotic use in 1 NICU and 2 NBNs. Despite automated data collection, regular data reporting to all providers was challenging because of the need for manual “refreshing” of the data, limited provider access to the dashboard, and competing priorities for the AS team. Because of competing priorities related to the coronavirus disease 2019 pandemic, further interventions and phase 2 of the project have not been implemented after the incorporation of the SRC in the EHR in August 2019.

This project successfully implemented an AS program at a community medical center, then scaled up the program to improve the quality of care for a 10-hospital system. The core elements of this program included education, data collection and sharing, a shortened sepsis rule-out period, and the implementation of the SRC. Starting with small pilots and sharing success stories, an opt-in strategy was used to scale the work. Hospitals with well-organized QI teams and smaller hospitals had fewer barriers to implementing these pilots. The ability to earn MOC4 credit was a significant motivator for most physicians involved. Although hospitals with multiple interventions saw the greatest drop in AUR, even hospitals with moderate involvement showed sustained drops in antibiotic use. By involving 8 hospitals in the system, we were able to leverage the health system resources to acquire IT support for long-term automated data collection and changes in the EHR that were instrumental in sustainability.

Thank you to Dr Atul Khurana, Dr Alan Glassman, Lindsay McCloskey, Sara Hughes, and many others for support at PAR. Many thanks to all the NICU directors and medical providers involved. Thank you to the Quality Improvement Department at PAR, as well as all the CMOs of the participating hospitals, for supporting the project financially. Thank you to Madge Buus-Frank for coaching and support. Thank you to Drs Stephen Balis, Caroline Geiger, Erin Shumans, Leah Williams, Courtney Alvis, Amanda Wachtel, and Raveena Patel, who all supported the work during their medical school careers. Thanks to Stephen Darkota and Ginny Hollifield for helping with the Tableau dashboard. Thank you to Ginny Hollifield, who took over the Tableau dashboard. Thank you to Lynn Ramsey and Jennifer Stowe for creating the figures for this article.

AS

antibiotic stewardship

AUR

antibiotic utilization rate

CDC

Centers for Disease Control and Prevention

EHR

electronic health record

iNICQ

Vermont Oxford Network quality improvement collaborative

IT

information technology

MOC4

maintenance of certification quality improvement credit

NBN

newborn nursery

PAR

Piedmont Athens Regional

PHC

Piedmont Healthcare

QI

quality improvement

SRC

neonatal early-onset sepsis risk calculator